Algebra Homotopy Operad

ALGEBRA + HOMOTOPY = OPERAD
BRUNO VALLETTE If I could only understand the beautiful consequence following from the concise proposition d2 = 0. Henri Cartan
A BSTRACT. This survey provides an elementary introduction to operads and to their applications in homotopical algebra. The aim is to explain how the notion of an operad was prompted by the necessity to have an algebraic object which encodes higher homotopies. We try to show how universal this theory is by giving many applications in Algebra, Geometry, Topology, and Mathematical Physics. (This text is accessible to any student knowing what tensor products, chain complexes, and categories are.)
C ONTENTS Introduction 1. When Algebra meets Homotopy 2. Operads 3. Operadic syzygies 4. Homotopy Transfer Theorem Conclusion References 1 2 8 18 33 40 40
I NTRODUCTION Galois explained to us that operations acting on the solutions of algebraic equations are mathematical objects as well. The notion of an operad was created in order to have a well dened mathematical object which encodes operations. Its name is a porte-manteau word, coming from the contraction of the words operations and monad, because an operad can be dened as a monad encoding operations. The introduction of this notion was prompted in the 60s, by the necessity of working with higher operations made up of higher homotopies appearing in algebraic topology. Algebra is the study of algebraic structures with respect to isomorphisms. Given two isomorphic vector spaces and one algebra structure on one of them, one can always dene, by means of transfer, an algebra structure on the other space such that these two algebra structures become isomorphic. Homotopical algebra is the study of algebraic structures with respect to quasi-isomorphisms, i.e. morphisms of chain complexes which induce isomorphisms in homology only. The main difference with the previous situation is that quasi-isomorphisms of chain complexes are not invertible in general. However, given two quasi-isomorphic chain complexes and one algebra structure on one of them, one can always dene, on the other chain complex, an algebra structure. But, in this case, the relations are satised only up to homotopy. In the end, these two algebra structures become quasi-isomorphic in some way. This last result is called the Homotopy Transfer Theorem, or HTT for short. In the rst cases like associative algebras, commutative algebras, and Lie algebras, this kind of result can be proved by hand. However, in general, the transferred algebra structure yields a coherent system of higher homotopies, whose complexity increases quickly with the number of the initial operations. Using operad theory, for instance Koszul duality theory, one can prove the HTT with explicit formulae. In this way, one recovers, and thus unies, many classical objects appearing in Algebra, Geometry, Topology and, Mathematical Physics like spectral sequences, Massey products, or Feynman diagrams, for instance.
1
BRUNO VALLETTE
This text does not pretend to be exhaustive, nor to serve as a faithful reference to the existing literature. Its only aim is to give a gentle introduction to the ideas of this eld of mathematics. We would like to share here one point of view on the subject, from student to student. It includes many gures and exercises to help the learning reader in its journey. To ease the reading, we skipped many technical details, which can be found in the book [LV10]. C ONVENTION . In this text, a chain complex (V, d) is usually a graded module V := {Vn }nZ equipped with a degree 1 map d (homological convention), which squares to zero. For the simplicity of the presentation, we always work over a eld K of characteristic 0, even if some exposed results admit generalizations beyond that case. 1. W HEN A LGEBRA MEETS H OMOTOPY In this section, we treat the mother example of higher algebras: A -algebras. We show how this notion appear naturally when one tries to transfer the structure of an associative algebra through homotopy data. We provide an elementary but extensive study of its homotopy properties (Homotopy Transfer Theorem, A -morphism, Massey products and homotopy category). 1.1. Homotopy data and algebraic data. Let us rst consider the following homotopy data of chain complexes:
h
(A, dA ) o
p i
/ (H, d ) H
IdA ip = dA h + hdA , where i and p are morphisms of chain complexes and where h is a degree +1 map. It is called a homotopy retract, when the map i is a quasi-isomorphism, i.e. when it realizes an isomorphism in homology. If moreover pi = IdH , then it is called a deformation retract. Since we work over a eld, any quasi-isomorphic chain complexes are related by a homotopy retract. Independently, let us consider the following algebraic data of a associative algebra structure on A: : A2 A, such that ( (a, b), c) = (a, (b, c)), a, b, c A .
By simplicity, we choose to depict these composites by the following graphically composition schemes:
where we forget about the input variables since they are generic. Actually, we consider a differential graded associative algebra structure on (A, dA ), dga algebra for short, on A. This means that the differential dA of A is a derivation for the product :
(In this section, we do not require that the associative algebra has a unit.) 1.2. Transfer of algebraic structure. One can now try to study how these data mix. Namely, is it possible to transfer the associative algebra structure from A to H through the homotopy data in some way ? The rst answer is given by the following formula for a binary product 2 := p i2 : H 2 H on H . Under our graphical representation, it gives
Now, let us check whether the product 2 is associative:
If the map i were an isomorphism, with inverse p, the answer would be yes. But, this does not hold in general; so the answer is no. Since the composite ip is equal to the identity up to the homotopy h, the transferred product 2 seems to be associative only up to the homotopy h. Let us make this statement precise. The associativity relation is equivalent to the vanishing of the associator
in Hom(H 3 , H ). This mapping space becomes a chain complex when equipped with the usual differential map (f ) := dH f (1)|f | dH 3 f . We introduce the element 3
It has degree +1, since it is the composite of maps of degree 0 (the maps i, p, and ) and one map of degree 1 (the map h). E XERCISE 1. We leave it to the reader to check that
This means that, in the chain complex (Hom(H 3 , H ), ), the associator of 2 is not zero in general, but vanishes up to the homotopy 3 . Under this property, one says that the product 2 is associative up to homotopy. The next step consists in checking whether the two operations 2 and 3 satisfy some relation. The answer is again yes, they satisfy one new relation but only up to yet another homotopy, which is a linear map 4 of degree +2 in Hom(H 4 , H ). And so on and so forth.
BRUNO VALLETTE
Let us cut short and give a general answer: generalizing the previous two formulae, we consider the following family of maps
n =
in Hom(H n , H ), for any n 2, where the notation P BTn stands for the set of planar binary rooted trees with n leaves. This denes linear maps of degree n 2, which satisfy the following relations
1.3. A -algebra. The aforementioned example triggers the following general denition. D EFINITION (A -algebra). An associative algebra up to homotopy, also called homotopy associative algebra or A -algebra for short, is a chain complex (A, d) endowed with a family of operations n : An A of degree n 2, for any n 2, satisfying the aforementioned relations. This notion was discovered by Jim Stasheff in his Ph.D thesis [Sta63] on loop spaces. E XAMPLE . An A -algebra, whose higher operations n = 0 vanish for n 3, is a dga algebra. If we dene a morphism of A -algebras to be a chain map commuting with the respective operations, then the category of dga algebras is a full sub-category of the category of A -algebras: dga alg A -alg . Theorem 1 (Homotopy Transfer Theorem for dga algebras [Kad80]). The operations {n }n2 dened on H from the associative product on A by the formulae of Section 1.2 form an A -algebra structure. Proof. We leave the proof to the reader as an easy and pedagogical exercise. 1.4. Application: Massey products. The rst kind of examples is given by considering (H, dH ) := (H (A, dA ), 0) together with the following deformation retract. First the degree n space An is the direct sum An = Zn Bn1 of the n-cycles Zn := {x An ; dA (x) = 0} with the (n 1)-boundaries Bn1 := dA (An ). Then, one lifts the homology groups Hn := Zn /Bn , to nally get An = Bn Hn Bn1 . Such a decomposition induces the required maps: i : Hn h : An 1 Bn Hn Bn1 , p : Bn Hn Bn1 Hn and B n1 B n Hn B n1 = Bn1 Hn1 Bn2 = An .
It is an easy exercise to prove that the product of a dga algebra induces an associative product on the underlying homology groups. Moreover, Theorem 1 allows us to endow H (A) with a much richer structure extending = 2 . Proposition 2. The underlying homology groups H (A) of a dga algebra A carry an A -algebra structure, with trivial differential, and extending the induced binary product.
D EFINITION (A -Massey products). We call A -Massey products the operations {n : H (A)n H (A)}n2 of the transferred A -algebra structure on H (A). In the light of the formulae of Section 1.2, we would like to say that the induced binary product on homology is associative for a bad reason: not that the rst homotopy 3 vanishes, but just because the differential map dH is trivial ! In this case, even if 2 is associative, the higher homotopies {n }n3 are not trivial, in general, as the example below shows. The A -Massey products actually encode faithfully the homotopy type of the dga algebra A, see Section 1.7. The classical Massey products, as dened by William S. Massey in [Mas58], are only partially dened and live on some coset of the homology groups. For example, the triple Massey product x , y , z is dened for homology classes x , y , z such that ( x, y ) = 0 = ( y, z ) as follows. Let us denote by x, y, z cycles representing the homology classes. Because of the hypothesis, there exist chains a and b such that (1)|x| (x, y ) = dA (a), (1)|y| (y, z ) = dA (b). Then the chain (1)|x| (x, b) + (1)|a| (a, z ) is a cycle and so denes an element x , y , z in H (A)/{ ( x, H (A)) + (H (A), z )}. E XERCISE 2. Show that 3 (x, y, z ), as dened by the formula of Section 1.2, provides a representative for the partial triple Massey product x , y , z . The formulae dening the Massey products are similar to the formulae dening the A -Massey products. The difference lies in the fact that, in the latter case, all the choices are made once and for all in the deformation retract. (The various liftings are then given by the homotopy h.) Proposition 3. The A -Massey products provide representatives for the partial Massey products. The original Massey products were dened in the context of the singular cohomology since, for any topo logical space X , the cup product endows the singular cochain complex Csing (X ) with a dga algebra structure, cf. [Hat02, Section 3.2]. A PPLICATION (Borromean rings). The Borromean rings are three interlaced rings, which represent the coat of arms of the Borromeo family in Italy, see Figure 1. Removing any one of the three rings, the remaining two rings are no longer interlaced. We consider the complement in the 3-sphere of the Booromean rings. Each ring, i.e. solid torus, determines a 1-cocycle: take a loop from the base-point with linking number one with the circle. Since any two of the three circles are disjoint, the cup product of the associated cohomology classes is 0. The nontriviality of the triple Massey product of these three cocycles detects the nontrivial entanglement of the three circles, cf. [Mas58].
F IGURE 1. Coat of arms of the Borromeo family
BRUNO VALLETTE
1.5. Homotopy Transfer Theorem. We have seen that starting from a dga algebra structure on one side of a homotopy data, one gets an A -algebra structure, with full higher operations, on the other side. What will happen if one starts from an A -algebra structure {n : An A}n2 ? In this case, one can extend the aforementioned formulae, based on planar binary trees, by considering the full set P Tn of planar rooted trees with n leaves, this time:
n =
Theorem 4 (Homotopy Transfer Theorem for A -algebras [Kad82]). The operations {n }n2 dened on H from the A -algebra structure {n }n2 on A by the aforementioned formulae form an A -algebra structure. Proof. Once again, we leave this straightforward computation to the learning reader. So, with the denition of A -algebras, we have reached a big enough category, including that of dga algebras, and stable under transfer through homotopy data. It remains to dene a good notion of morphism. 1.6. A -morphism. D EFINITION (A -morphism). An A -morphism between two A -algebras (A, dA , {n }n2 ) and (B, dB , {n }n2 ) is a collection of linear maps {fn : An B }n1 of degree n 1 satisfying
where 1 = dA and 1 = dB . The data of an A -morphism is denoted f : A A -morphisms f : A B and g : B C is dened by
B . The composite of two
(g f )n :=
E XERCISE 3. Show that A -algebras with A -morphisms form a category. We denote this category by -A -alg, where the rst symbol stands for the A -morphisms. By denition, the rst component f1 : A B of an A -morphism is a morphism of chain complexes. D EFINITION (A -isomorphism and A -quasi-isomorphism). When the map f1 is an isomorphism (resp. a quasi-isomorphism), the A -morphism f is called an A -isomorphism (resp. an A -quasi-isomorphism). E XERCISE 4. Show that A -isomorphisms are the invertible morphisms of the category -A -alg. (This proof is similar to the proof that power series a1 x + a2 x2 + with rst term invertible are invertible). Theorem 5 (Homotopy Transfer Theorem with A -quasi-isomorphism [KS01]). Starting from a homotopy retract, the maps 1 := i : H A and
n :=
: H n A ,
for n 2, dene an A -quasi-isomorphism :H A between the transferred A -algebra structure {n }n2 on H and the initial A -algebra structure {n }n2 on A. Proof. As usual, the proof is left to the reader as a good exercise. One can actually proceed one step further and extend the whole homotopy data from the category of chain complexes to the category of A -algebras. There exists a notion of an A -homotopy between A morphisms, whose equivalence relation is denoted h . Starting from a deformation retract, the complete result states that the quasi-isomorphism p extends to an A -quasi-isomorphism p and that the homotopy h extends to an A -homotopy between IdA and the composite p , cf. [Mar04]. 1.7. Homotopy theory of A -algebras. Theorem 6 (Fundamental theorem of A -quasi-isomorphisms [LH03]). Any A -quasi-isomorphism A B admits a homotopy inverse A -quasi-isomorphism B A.
This is the main property of A -quasi-isomorphisms. Notice that given a quasi-isomorphism of dga algebras, there does not always exist a quasi-isomorphism in the opposite direction as dga algebras. These two notions of quasi-isomorphisms are related by the following property: there exists a zig-zag of quasiisomorphisms of dga algebras if and only if there exists a direct A -quasi-isomorphism A B

B.
D EFINITION (Formality). A dga algebra (A, d, ) is called formal if there exists a zig-zag of quasiisomorphisms of dga algebras (A, d, ) (H (A), 0, ) . Proposition 7. (1) If a dga algebra is formal, then the higher Massey products, i.e. for n 3, vanish. (2) If the higher A -Massey products, i.e. for n 3, vanish, then the dga algebra is formal. Proof. The proof of the second point is a corollary of the HTT, Theorem 5. When the higher A -Massey products vanish, the map : H (A) A is an A -quasi-isomorphism between two dga algebras. We conclude then with the above property. In the latter case, the Massey products vanish in a uniform way, see [DGMS75] for more details.

BRUNO VALLETTE
D EFINITION (Homotopy category). The homotopy category of dga algebras Ho(dga alg) := dga alg[qi1 ] is dened as the localization of the category of dga algebras with respect to the class of quasi-isomorphisms. This is the smallest category containing the category dga alg in which quasi-isomorphisms become invertible. It is actually made up of more maps than the category dga alg since the maps of the homotopy category are (some equivalence classes) of chains of maps either from the morphisms of dga algebras or from the formal inverse of quasi-isomorphisms: /h / / /h / h / /B. / A

The following results shows that the homotopy classes of A -morphisms correspond to the maps in the homotopy category of dga algebras. Theorem 8 (Homotopy category [Mun78, LH03]). The following categories are equivalent Ho(dga alg) = -A alg/ h = -dga alg/ h . Proof. The last equivalence with the category of dga algebras equipped with the A -morphisms is given by the Rectication property: any A -algebra admits a dga algebra, which is A -quasi-isomorphic to it. Q UESTIONS . (1) Why this particular denition of an A -algebra when starting from an associative algebra ? (2) How could we perform the same theory starting from other types of algebras like commutative algebras, Lie algebras or Lie bialgebras, for instance ? 2. O PERADS In this section, we introduce the notion of an operad trough the paradigm given by the multilinear maps of a vector space. One can also use operads in the context of topological spaces, differential graded modules and simplicial sets, for instance. Operads encode categories of algebras having operations having multiple inputs and one output. When one wants to encode other types of algebraic structures, one has to introduce some generalizations of operads. 2.1. Unital associative algebras and their representation theory. Let V be a vector space. The vector space Hom(V, V ) of endomorphisms of V , equipped with the composition of functions and the identity morphism idV , becomes a unital associative algebra. Representation theory is the study of morphisms of unital associative algebras with target the endomorphism algebra: : (A, , 1) (Hom(V, V ), , idV )
This means that the map sends the unit 1 to the identity idV and any product of two elements of A to the associated composite
((b a)) = (b) (a) =
On the one hand, the endomorphism space Hom(V, V ) is the space of all concrete linear operations acting on V . On the other hand, the elements composing the image of the representation form a particular choice of operations. The type (number, relations, etc.) of these operations is encoded, once and for all V and , by the algebra A. E XAMPLE (Algebra of dual numbers and chain complexes). Let us begin with an elementary example. Suppose that we want to encode in one associative algebra, the algebraic data of a unary square-zero operator. In this case, we would rst consider the free unital associative algebra on one generator : it is explicitly given by the tensor module T (K ) := (K )n ,
n N
equipped with the concatenation product. This is meant to represent all the possible formal compositions of the operator with itself. Any morphism of unital associative algebras : T (K ) Hom(V, V ) is completely characterized by the image d := ( ) of this generator. Then we quotient this free algebra by the ideal generated by 2 , that is the relation we want to model. It gives the algebra of dual numbers: D := T (K )/( 2 ) . In this case, a morphism of unital associative algebras : D Hom(V, V ) is characterized by the image d := ( ) of the generator , which has to satisfy the relation d2 = 0. Finally, a vector space equipped with a square-zero operator is a representation of D. If we work with graded vector spaces and if we place the generator in degree 1, then we get nothing but the notion of a chain complex. E XAMPLE (Steenrod operations and Steenrod algebra). Even if the previous example can seem naive at rst sight, the idea of abstracting families of unary operations, which act in the same way is not trivial. Consider again the singular cohomology Hsing (X ) of topological spaces. Norman E. Steenrod constructed in [Ste47] natural unary operations, the Steenrod squares {Sq i }i1 in characteristic 2, which act on the singular cohomology Hsing (X ). They are higher operators which come from the higher homotopies for the commutativity of the cup product. They are functorial and always satisfy the same relations, the Adem relations, for any topological space X . So, Henri Cartan introduced in [Car55] an algebra A2 := T ({Sq i }i1 )/(RAdem ) naturally called the Steenrod algebra. It encodes all the Steenrod squares at once. Proving a functorial result for all Steenrod operations amounts to proving only one statement on the level of the Steenrod algebra. 2.2. Operads and multilinear representation theory. We aim now at developing the same kind of theory but with operations having multiple inputs this time. On the right hand side, we consider the family EndV := {Hom(V n , V )}n0 of multilinear maps. The goal now is to nd what kind of mathematical object P should be put on the left hand side: : P ? EndV . Let us extract the properties of EndV as follows. (i) First, it is made of a family of vector spaces indexed by the integers, so will be P := {Pn }nN . (ii) Then, the space EndV of multilinear maps is equipped with the classical composition of functions. Given a rst level of k operations and one operation with k inputs, one can compose them in the following way
10
BRUNO VALLETTE
Mimicking this composition, we require P to be equipped with composite maps: i1 ,...,ik : Pk Pi1 P ik 1 k Pi1 ++ik (; 1 , . . . , k )
that we of course represent similarly. The composition of functions is associative in the following way: if we consider 3 levels of functions, composing the two upper levels and then the result with the third one or beginning with the two lower levels and then compositing with the rst one, this gives the same function. This associativity condition reads as follows.
(iii) Finally, we have the identity morphism idV in Hom(V, V ). So, we ask for a particular element I in P (1), which behaves as a unit for the composite maps: 1,...,1 : I I and k : I . D EFINITION (Nonsymmetric operad). A nonsymmetric operad P , ns operad for short, is a family {Pn }nN of vector spaces with an element I P1 and endowed with associative and unital composite maps i1 ,...,ik : Pk Pi1 Pik Pi1 ++ik . E XAMPLES . The endomorphism operad EndV := {Hom(V n , V )}n0 is the mother of (nonsymmetric) operads. Let (A, , 1A ) be a unital associative algebra. We consider the family A dened by A1 := A and An := 0 for n = 1 and we set I := 1A . All the composite maps are equal to zero, except for 1 := . So a unital associative algebra induces a ns operad. In the other way round, any ns operad concentrated in arity 1 is a unital associative algebra. Since the notion of a nonsymmetric operad is a generalization of the notion of unital associative algebra, we will be able to extend many results from associative algebras to operads. D EFINITION (Morphism of ns operads). A morphism f : P Q between two ns operads (P , , IP ) and (Q, , IQ ) is a family {fn : Pn Qn }n0 of linear morphisms which preserve the units f1 (IP ) = IQ and which commute with the composite maps i1 ,...,ik (fk fi1 fik ) = fi1 ++ik i1 ,...,ik . We now dene the representations of a nonsymmetric operad P in a vector space V by the morphisms of nonsymmetric operads: : P EndV . Such a map associates to any formal operation Pn a concrete n-multilinear operation () Hom(V n , V ).
11
The operations in the image of satisfy the same relations as that of P . In the other way round, given a certain category of algebras, like associative algebras, commutative algebras or Lie algebras, for instance, we are able to encode the space of all possible operations into one operad, as follows. D EFINITION (P -algebra). Let P be an operad. A P -algebra structure on a vector space V is a morphism of ns operads : P EndV . 2.3. The examples of As and uAs. Let us see how this works on an example. We go back to the example of Section 1: the category of associative algebras, not necessarily unital. Let us now try to gure out the operad, which models this category. Given n elements a1 , . . . , an of an associative algebra A, what are the operations, induced by the associative binary product, acting on them ? Keeping the variables in that order, that is without permuting them, there is just one because any bracketing will produce the same result under the associativity relation. Hence, we decide to dene
Asn := K the one-dimensional vector space, for any n 1. For an algebra over an operad, the set P0 is sent to Hom(K, V ), since V 0 = K by convention. Therefore it stands for the set of particular elements in V . In the case of associative algebras, we have none of them. There is no unit for instance; so we dene As0 := 0. The composite map i1 ,...,ik : Ask Asi1 Asik Asi1 ++ik is the multiplication of scalars, i.e. the isomorphism K K K = K. E XERCISE 5. Show that, with this denition, the data (As, , I) forms a nonsymmetric operad. The next proposition shows that we have done a good job. Proposition 9. The category of As-algebras is isomorphic to the category of associative algebras. Proof. Let : As EndV be a representation of the ns operad As. Since As0 = 0 and As1 = K I, c c c ). It denes a binary product on V . The by := (c they code for c nothing. Let us call the image of c c c c c c c composite of ( I ) and (I ) in As give the same result. Therefore the composite of their respective image in Hom(V 3 , V ) are equal, which gives the associativity of . As usual, we leave the rest of the proof to the reader as a good exercise. If we want to model the category of unital associative algebras this time, we dene the operad uAs in the same way, except for uAs0 := K 1. E XERCISE 6. Show that (uAs, , I) forms a nonsymmetric operad and that uAs-algebras are unital associative algebras. R EMARK . At this point of the theory, some readers might be confused by the fact that unital associative algebras appeared twice: on the one hand, the notion of a nonsymmetric operad is a generalization of the notion of unital associative algebra and, on the other hand, there is one operad uAs, which encodes the category of unital associative algebras. So, one should be careful. In these two cases, the category of unital associative algebras does not play the same role and is not placed on the same footing. 2.4. Symmetric operads. Let us now extend the denition of a nonsymmetric operad in order to take care of the possible symmetries of the operations that we aim to encode. First, we notice that the space Hom(V n , V ) of n-multilinear maps carries a natural right action of the symmetric group Sn , induced by the permutation of the input elements: f (v1 , . . . , vn ) := f (v1 (1) , . . . , v1 (n) ) . The composition of multilinear functions satises natural equivariance properties with respect to this action. E XERCISE 7. Write down these equivariance properties.
12
BRUNO VALLETTE
D EFINITION (Symmetric operad). A symmetric operad P , or simply an operad, is a family {P (n)}nN of right Sn -modules with an element I P (1) and endowed with associative, unital and equivariant composite maps i1 ,...,ik : P (k ) P (i1 ) P (ik ) P (i1 + + ik ) . We refer to this denition as to the classical denition of an operad, since it coincides with the original definition of J. Peter May in [May72]. The denition of a morphism of symmetric operads and the denition of an algebra over a symmetric operad are the equivariant extensions of the nonsymmetric ones. So, forgetting the action of the symmetric groups, one denes a functor from symmetric operads to nonsymmetric operads. E XERCISE 8 (Partial denition). Show that the denition of an operad structure on a family of Sn -modules {P (n)}n0 is equivalent to the data of partial compositions i : P (m) P (n) P (m 1 + n), for 1 i m, satisfying the equivariance with respect to the symmetric groups (to be made precise), the axioms: (I) (II) ( i ) i1+j ( i ) k1+m = = i ( j ), for 1 i l, 1 j m, ( k ) i , for 1 i < k l,
for any P (l), P (m), P (n), and having a unit element I P (1). We refer to this denition as to the partial denition of an operad. E XAMPLES . Let us dene the symmetric operad Com (respectively uCom), in the same way as the nonsymmetric operad As (respectively uAs), by the one-dimensional spaces
Com(n) := K but endowed with the trivial representation of the symmetric groups, this time. As usual, we leave it, as a good exercise, to the reader to check that these data form an operad. Prove also that the category of Com-algebras (respectively uCom-algebras) is isomorphic to the category of commutative algebras (respectively unital commutative algebras). In contrast to the operad Com of commutative algebras, the various known bases of the space of n-ary operations of the operad Lie of Lie algebras do not behave easily with respect to both the operadic composition and to the action of the symmetric groups. n On the one hand, we know that the right module Lie(n) is the induced representation IndS Z/nZ (), where () is the one-dimensional representation of the cyclic group given by an irreducible nth root of unity. On the other hand, the following bases of Lie(n), due to Guy Melanc on and Christophe Reutenauer in [MR96], behaves well with respect to the operadic composition. We consider the subset M Rn of planar binary trees with n leaves such that, at any given vertex, the left-most (resp. right-most) index is the smallest (resp. largest) index among the indices involved by this vertex. Here are the elements of M R2 and M R3 .
13
We dene the partial compositions t i s of two such trees by grafting the root of s on the leaf of t with label i and by shifting the indices accordingly. We leave it to the reader to verify that this endows {K M Rn } with a nonsymmetric operad structure. For more properties about this basis of the operad Lie, we refer the reader to the end of Section 3.5. E XERCISE 9. One can also encode the category of associative algebras with a symmetric operad that we denote Ass, where the second s stands for symmetric. Since the multilinear operations of an associative algebra have in general no symmetry, the component of arity n of the operad Ass is the regular representation of the symmetric group: Ass(n) := K[Sn ]. We leave it to the reader to make the composite maps explicit. 2.5. Changing the underlying monoidal category. One can also consider operads not only over the category of vector spaces but over other categories as follows. In order to dene the notion of an operad, we used the fact that the category (Vect, ) of vector spaces is a monoidal category, when endowed with the tensor product . To write down the associativy property of the composite maps , one needs this monoidal category to be symmetric because we have to commute terms separated by tensors. In the end, one can dene the notion of an operad over any symmetric monoidal category. To name but a few:
S YMMETRIC MONOIDAL CATEGORY T YPE OF OPERAD Vector spaces (Vect, ) Linear operads Graded modules (gr Mod, ) Graded operads Differential graded modules (dg Mod, ) Differential graded operads Sets (Set, ) Set-theoretic operads Topological spaces (Top, ) Topological operads Simplicial sets (Set , ) Simplicial operads
Set-theoretical operad: the example of monoids. Let us dene the set-theoretic ns operad M on, in the same way as the ns operad uAs, by the following one-element sets:
M onn :=
for n N. In this case, the category of M on-algebras is isomorphic to the category of monoids.
Topological operad: the example of the little discs. The mother of topological operads is the little discs operad D2 , which is a topological operad dened as follows. The topological space D2 (n) is made up of the unit disc (in C) with n sub-discs with non-intersecting interiors. So, an element on D2 (n) is completely determined by a family of n continuous maps fi : S 1 D2 , i = 1, . . . , n, satisfying the non-intersecting condition, see Figure 2.
14
BRUNO VALLETTE
F IGURE 2. Little discs conguration in D2 (3) The enumeration of the interior discs is part of the structure. The operadic composition is given by insertion of a disc in an interior disc. The symmetric group action is given by permuting the labels. Figure 3 gives an example of a partial composition.
= F IGURE 3. Exemple of partial composition in the little discs operad It is clear how to proceed to dene the little k -discs operad Dk , for any k N. In the case k = 1, one gets the little intervals operad. E XERCISE 10. Prove that any k -fold loop space k (Y ) is an algebra over Dk . The main property of the little k -discs operad is the following result, which goes the other way round. Theorem 10 (Recognition principle [BV73, May72]). If the connected topological space X is an algebra over the little k -discs operad, then it is homotopy equivalent to the k -fold loop space of some other pointed space Y : X k (Y ).
15
So, up to the homotopy relation of topological spaces, the category of Dk -algebras is the category of k -fold loop spaces. Differential graded operad: the example of the Gerstenhaber operad. A Gerstenhaber algebra is a graded vector space A endowed with a commutative binary product of degree 0, a commutative bracket , of degree +1, i.e. | a, b | = |a| + |b| + 1, where the notation | | stands for the grading of elements. such that the product is associative, the bracket satises the Jacobi identity , , + , , .(123) + , , .(321) = 0,
the product and the bracket , satisfy the Leibniz relation -, - = ( -, - -) + (- -, - ).(12), We denote by Ger the graded symmetric operad which encodes Gerstenhaber algebras. Any symmetric monoidal functor between two symmetric monoidal categories induces a functor between the associated notions of operads. For instance, the free vector space functor (Set, ) (Vect, ) is symmetric monoidal. Under this functor, the set-theoretic ns operad M on is sent to the linear ns operad uAs = KM on. Since we are working over a eld, the homology functor H : (Top, ) (gr Mod, ) , denes such a symmetric monoidal functor. Theorem 11 (Homology of the little discs operad [Coh76]). The homology of the little discs operad is isomorphic to the operad encoding Gerstenhaber algebras H (D2 ) = Ger . 2.6. Other types of Operads. Depending on the type of algebraic structure that one wants to encode (several spaces, scalar product, trace, operations with multiple inputs and multiple outputs, etc.), it is often possible to extend the previous denition of an operad. Colored operad. Algebraic structures acting on several underlying spaces can be encoded with colored operads. In this case, the paradigm is given by EndV1 Vk . An operation in a colored operad comes equipped with a color for each input and a color for the output. The set of colors corresponds to the indexing set of the various spaces, for instance {1, . . . , k }. E XERCISE 11. The data of two associative algebras (A, A ) and (B, B ) with a morphism f : A B between them can be encoded into a 2-colored operad denoted As . Show that this operad is made up of corollas of three types : only black inputs and output (for the source associative algebra A), only white inputs and output (for the target associative algebra B ), and corollas with black inputs and white output (either for the composite of iterated products in A and then the morphism f or for the composite of tensors of copies of the morphism f and then iterated products in B ).
E XAMPLE . An interesting example of topological colored operad is given by the operad of tangles of Vaughan F.R. Jones [Jon10], see Figure 4. (Notice the similarity with the little discs operad.) Algebras over this operad are the planar algebras [Jon99], notion related to Jones invariants of knots.
16
BRUNO VALLETTE
F IGURE 4. A Tangle (Courtesy of V.F.R. Jones)
Cyclic operad. The purpose of cyclic operads is to encode algebraic structures acting on spaces endowed with a non-degenerate symmetric pairing , : V V K. In this case, the endomorphism operad EndV (n) = Hom(V n , V ) = Hom(V n+1 , K) is equipped with a natural action of Sn+1 , which extends the action of Sn and which is compatible with the partial compositions. On the level of cyclic operads, this means that we are given an action of the cycle (1 2 . . . n + 1) which exchanges input and output:
As usual, the data of an algebra over a cyclic operad is given by a morphism of cyclic operads P EndV . For example an algebra over the cyclic operad Ass is a nite dimensional associative algebra equipped with a non-degenerate symmetric pairing , , which is invariant: ab, c = a, bc . E XAMPLE . A Frobenuis algebra, central notion in representation theory [CR06], is nothing but a cyclic uAs-algebra. (The nonsymmetric operad uAs-can be endowed with a cyclic nonsymmetric operad structure, that is with an action of the cyclic groups Z/(n +1)Z). A symmetric Frobenius algebra, central notion in Topological Quantum Field Theory [Koc04], is a cyclic uAss-algebra. Let us recall that a Frobenius algebra (resp. symmetric Frobenius algebra) is a unital associative algebra, equipped with a nondegenerate (resp. and symmetric) bilinear form, such that ab, c = a, bc . E XERCISE 12. Show that the operad Lie can be endowed with a cyclic operad structure. Prove that any nite dimensional Lie algebra g, becomes an algebra over the cyclic operad Lie, when equipped with the Killing form. Modular operad. In a cyclic operad, one makes no distinction between inputs and outputs. So one can compose operations along genus 0 graphs. The idea of modular operad is to also allow one to compose operations along any graph of any genus. E XAMPLE . The Deligne-Mumford moduli spaces Mg,n+1 [DM69] of stable curves of genus g with n + 1 marked points is the mother of modular operads. The operadic composite maps are dened by intersecting curves along their marked points:
17
The Gromov-Witten invariants of enumerative geometry endow the cohomology groups H (X ) of any projective or symplectic variety with a H (Mg,n+1 )-algebra structure, see [KM94]. A Frobenius manifold [Man99] is an algebra over the cyclic operad H (M0,n+1 ). By denition, the Quantum cohomology ring is the H (M0,n+1 )-algebra structure on H (X ). Properad. A properad {P (n, m)}m,nN is meant to encode operations with several inputs and several outputs. But, in contrast to modular operads, where inputs and outputs are confused, one keeps track of the inputs and the outputs in a properad. Moreover, we consider only compositions along connected graphs as follows.
E XAMPLE . Riemann surfaces, i.e. smooth compact complex curves, with parametrized holomorphic holes form a properad. We denote by Rg,n,m its component of genus g with n input holes and m output holes. The properadic composite maps are dened by sewing the Riemann surfaces along the holes.
A Conformal Field Theory, as dened by Graeme Segal in [Seg04], is nothing but an algebra over the properad Rg,n,m of Riemann surfaces. In the rest of the text, we will introduce algebraic properads to encode various categories of bialgebras. The name properad is a porte-manteau word from prop and operad. Prop. A prop is like a properad, but where one can also compose along non-necessarily connected graphs. This is the operadic notion which was introduced rst, by Saunders MacLane in [ML65] as a symmetric monoidal category C, whose objects are the natural numbers and whose monoidal product is their sum. In
18
BRUNO VALLETTE
this case, the elements of HomC (n, m) are seen as the operations with n inputs and m outputs. An algebra over a prop is what F.-W. Lawvere called an algebraic theory in [Law63]. E XAMPLE . The categories of cobordism, where the objects are the d-dimensional manifolds and where the morphisms are the (d + 1)-dimensional manifolds with d-dimension boundary, form a prop. Sir Michael F. Atiyah proposed to dene a Topological Quantum Field Theory, or TQFT for short, as an algebra over a category of cobordism [Ati88]. Wheeled properad, wheeled prop. Sergei A. Merkulov introduced in [Mer09] the notion of wheels to encode algebras equipped with traces. A trace is a linear map tr : Hom(V, V ) K satisfying tr(f g ) = tr(gf ). We translate this on the operadic level by adding contraction maps:
i j : P (n, m)
P (n 1, m 1)
satisfying
There are obvious functors between these notions. For instance, xing an output, the genera 0 and 1 composition maps of a modular operad induce a wheeled operad structure. In general, a modular operad induces a properad, which contains all the higher genera compositions. The advantage is that properads can act on innite dimensional vector spaces and that a Koszul duality theory is dened on properads level, see 3.4. Now that we have opened the Pandoras box, we hope that the reader feels comfortable enough to go on and dene new kinds of operads, which suit its problem best. 3. O PERADIC SYZYGIES Using the notion of an operad, we can answer the two questions raised at the end of Section 1. The example studied in Section 1 shows that some categories of algebras over an operad are not stable under homotopy data: the category of associative algebras is not stable, but the category of A -algebras is stable, as Theorem 4 shows. The conceptual explanation is that the ns operad A , encoding A -algebras, is free. So, the method to provide a good framework for homotopy theory with a given category of algebras is as follows. First, one encodes the category of algebras by an operad P . Then, one tries to nd a free operad P , which resolves the operad P , that is which is quasi-isomorphic to it. Finally, the category of P -algebras carries
19
the required homotopy properties and the initial category of P -algebras sits inside it. operad P o
P = (T (X ), d) : free resolution / category of homotopy algebras
category of algebras
3.1. Quasi-free operad. As well as for unital associative algebras, operads admit free operads as follows. Forgetting the operadic compositions, one denes a functor U from the category of operads to the category of families of right modules over the symmetric groups Sn , called S-modules. D EFINITION (Free operad). The left adjoint T to the forgetful functor U : Operads S-modules : T produces the free operad, since it freely creates operadic compositions from any S-modules. E XERCISE 13. Show that the free operad T (X ) on an S-module X is equivalently characterized by the following universal property: there exists a natural morphism of S-modules X T (X ), such that any morphism of S-modules f : X P , where P is an operad, extends to a unique morphism of operads : T (X ) P : f / T (X ) Xi ii ii ii f f i i" P. One denes the same notion on the level of nonsymmetric operads by replacing the category of S-modules by the category of families of K-modules, called N-modules. Let us now make this functor explicit. To begin with the nonsymmetric case, let X be an N-module and let t be a planar tree. We consider the space t(X ) dened by trees of shape t with vertices labelled by elements of X according to their arity.
Its structure of vector space is given by identifying t(X ) with X k , where k is the number of vertices of the tree t. We dene the N-module T (X ) by the sum of all the X -labelled planar trees: T (X )n :=
tP Tn
t(X ) .
The operadic compositions are given by the grafting of trees. In the symmetric case, one considers the set of rooted trees in space, instead of planar rooted trees, with leaves labelled bijectively by 1, . . . , n. Then one labels the vertices with the elements of the S-module X in a coherent way with respect to the action of the symmetric group on X and on the set of inputs of the vertices, see [LV10, Section 5.5] for more details. In a similar way, one can perform free operad constructions over the various underlying monoidal categories mentioned in Section 2.5. Proposition 12 (Free operad). The tree-module T (X ), with the grafting of trees as operadic compositions, is the free operad on X . Proof. As usual, we leave this proof to the reader as a good exercise.
20
BRUNO VALLETTE
According to Section 2.5, a differential graded operad, dg operad for short, is a differential graded S-module (P , d) endowed with operadic compositions, for which the differential d is a derivation:
k
d( (; 1 , . . . , k )) = (d(); 1 , . . . , k ) +
i=1
(; 1 , . . . , d(i ), . . . , k ) .
E XERCISE 14. Show that the differential d on a free operad T (X ) is completely characterized by the image of the generators X T (X ). Make explicit the full differential d from its restriction to the space of generators. D EFINITION (Quasi-free operad). A dg operad whose underlying graded operad is free, that is after forgetting the differential, is called a quasi-free operad. The operad itself is free but not the differential map. In a free dg operad, the differential is completely characterized by the differential of the space of generators dX : X X and carries no more terms. In a quasi-free operad, the differential associates to a generator of X a sum of labelled trees in T (X ), as the example of the next section shows. 3.2. Operadic syzygies. Let us make explicit a quasi-free resolution for the ns operad As. We shall begin with a topological resolution of the set-theoretic ns operad M on dened by M on0 := , and M onn := M onn = , for n 1 , and which provides a basis for As. In arity 1, the ns operad M on has one element of degree 0, so as the free ns operad T (X ) on a generating set X empty in arity 1: X1 = . Hence, in arity 1, one has two 0-cells, M on1 = T (X )1 = {I}, which are obviously homotopy equivalent.
In arity 2, the ns operad M on has one element
of degree 0. So, we introduce a degree 0 element
of arity 2 in the generating set X of T (X ). In arity 2, one has now two 0-cells, which are homotopy equivalent.
In arity 3, on the one hand, the ns operad M on has one element of degree 0: the free ns operad T (X ) has 2 different elements of degree 0: the left comb
. On the other hand, and the right comb .
So, we introduce in X3 a degree 1 element of arity 3, whose boundary is equal to these two degree 0 elements respectively. Finally, in arity 3, one has now one 0-cell, on one side, and two 0-cells linked by a 1-cell, on the other side. These two topological spaces are homotopy equivalent.
In arity 4, on the one hand, the ns operad M on has one element of degree 0. On the other hand, the operad T (X ) has 5 elements of degree 0, the 5 planar binary trees with 4 leaves, and 5 elements of degree 1, the 5 planar trees with 2 vertices and 4 leaves. They are attached to one another to form a pentagon.
21
To get a topological space, homotopy equivalent to the point, we introduce in X4 a degree 2 element , whose boundary is equal to the boundary of the pentagon. This kills the homology group of degree 1 in the free operad. E XERCISE 15. Proceed in the same way in arity 5 to discover the following polytope of dimension 3 with 14 0-cells, 20 1-cells, 9 2-cells, and 1 new 3-cell. Label it !
In general, one has to introduce one cell of dimension n 2 in arity n to kill the (n 2)th -homotopy group. In the end, the generating space X is made up of one element of degree n 2 in arity n, for n 2. The boundary of each generator is given, according to some orientation, by the set of planar trees with 2 vertices. These polytopes are called the associahedra, or the Stasheff polytopes after Jim Stasheff [Sta63]. (They also correspond to the Hasse diagrams of the Tamari lattices [Tam51] made up of planar binary trees with the partial order generated by the covering relation operad to get a topological resolution of M on.) D EFINITION (Operad A ). We dene the dgns operad
cell A := (C (T (X )), d)
. We could also use the little intervals
by the image, under the cellular chain functor, of the aforementioned free topological resolution T (X ) of the ns operad M on. Theorem 13 (Resolution A ). The dgns operad A provides the following quasi-free resolution of As: A = T , , ,... ,d As
where the boundary condition for the top dimensional cells translates into the following differential:
22
BRUNO VALLETTE
Proof. Since the topological ns operad T (X ) is homotopy equivalent to M on, the dgns operad A is quasi-isomorphic to As. Last step of the method explained in the introduction: an algebra over the resolution A of As is an A -algebra, as dened in 1.3. Indeed, a morphism of dgns operads A = T , , , . . . , d (EndV , )
in characterized by the images of the generators, which give here a family of operations n : V n V of degree n 2, for any n 2. The commutativity of the differentials applied to the generators amounts precisely to the relations given in 1.3 of an A -algebra. Finally, this section answers Question (1) raised at the end of Section 1. D EFINITION (Operadic syzygies). The generating elements X of a quasi-free resolution T (X ) are called the operadic syzygies. The problem of nding a quasi-free resolution for an operad is similar to the problem of nding a quasi-free resolution A M2 A M1 A M0 M for a module M over a commutative ring A in algebraic geometry, see [Eis05]. In this case, the generating elements {Mn }n0 of the free A-module form the classical syzygies. As in the classical case of modules over a commutative ring, we have introduced here a rst syzygy corresponding to the generator of As. Then, we have introduced a second syzygy for the relation of As. The third syzygy was introduced for the relations among the relations, etc. This iterative process is called the Koszul-Tate resolution after Jean-Louis Koszul [Kos50] and John Tate [Tat57]. (It was successfully used to compute the BRST cohomology in Mathematical Physics [HT92] and to compute rational homotopy groups in Rational Homotopy Theory [Sul77].) R EMARK . One can develop the homotopy theory for dg operads by endowing this category with a model category structure, see [Qui67, Hin97, BM03]. The conceptual reason why the category of algebras over a quasi-free operad behaves well with respect to homotopy data is that quasi-free operads are cobrant, the general notion for being projective. (This is actually true over non-negatively graded chain complexes. In general, one has to require the existence of a good ltration on the syzygies standing for the generators, the relations, the relations among the relations, etc.) The Koszul-Tate method works step-by-step. We would like now to be able to produce all the syzygies at once. 3.3. Ideal, quotient operad and quadratic operad. As well as for unital associative algebras, operads admit ideals and quotient operads. D EFINITION (Operadic ideal). A sub-S-module I P of an operad is called an operadic ideal if the operadic compositions satisfy (; 1 , . . . , k ) I , when at least one of the , 1 , . . . , k is in I . E XERCISE 16. Show that, in this case, the quotient S-module P /I carries a unique operad structure which makes the projection of S-modules P P /I into an operad morphism.
23
For any sub-S-module R of an operad P , there exists a smallest ideal containing R. It is called the ideal generated by R and denoted (R). The free operad T (E ) admits a weight grading T (E )(k) given by the number of vertices of the underlying tree. D EFINITION (Quadratic data and quadratic operad). A quadratic data (E ; R) consists of an S-module E and a sub-S-module R T (E )(2) . It induces the quadratic operad P (E ; R) := T (E )/(R) . E XAMPLES . The algebra of dual numbers D = T (K )/( 2 ) of Section 2.1 is a quadratic algebra, and thus a quadratic operad concentrated in arity 1. The ns operad As admits the following quadratic presentation As =P ; , ) resumes to planar binary trees and,
with one binary generator. Indeed, the free ns operad T ( under the relation among the subtrees of leaves identify to one another.
, the planar binary trees with the same number
In the same way, the operad Com admits the following quadratic presentation Com =P ; , ,
with one binary generator invariant under the action of S2 . The operad Lie is dened by the quadratic data Lie := P ; .
The presentation of the operad Ger encoding Gerstenhaber algebras, given in Section 2.5, is quadratic. In the same way, there exists a notion of free properad, explicitly given by labelled directed connected graphs. Properadic ideals and quotient are dened similarly, see [Val07] for more details. The category of Lie bialgebras, introduced by Vladimir Drinfeld in [Dri86], is encoded by the following quadratic properad BiLie := P (E ; R), where E := ,
with skew-symmetric action of S2 in both cases, and where R := ,
24
BRUNO VALLETTE
An involutive Lie bialgebra is a Lie bialgebra satisfying the following extra diamond condition: . Such a structure can be found in String Topology, see [CS99]. The properad BiLie encoding involutive Lie bialgebras is quadratic. In any symmetric Frobenius algebra, one can dualize the product via the non-degenerate bilinear form to dene a cocommutative coproduct. This gives rise to the notion of Frobenius bialgebra, which is encoded by the following quadratic properad: F rob := P , ; , , , .
E XERCISE 17. Show that the properad F rob is graded by the genus of the underlying graphs. Prove that the component F robg (n, m) of genus g , with n inputs and m outputs, is one dimensional. Then, make the properadic composition maps explicit. An involutive Frobenius bialgebra is a Frobenius bialgebra satisfying the extra diamond condition: . The associated properad admits the following quadratic presentation : F rob := P , ; , , , , .
E XERCISE 18. Show that in the properad F rob , the composite of operations along graphs of genus greater or equal than 1 vanishes. Prove that the component F rob (n, m) with n inputs and m outputs is one dimensional. Then, make the properadic composition maps explicit. D EFINITION (Quadratic-linear operad). A quadratic-linear operad is a quotient operad P (E ; R) := T (E )/(R) generated by a quadratic-linear data (E ; R): R E T (E )(2) . E XAMPLES . The universal enveloping algebra U (g) := T (g)/(x y y x [x, y ]) of a Lie algebra g is a quadratic-linear algebra. The Steenrod algebra A2 is the quadratic-linear algebra A2 := A({Sq i }i1 , RAdem ) over the characteristic 2 eld F2 , where |Sq i | = i and where RAdem stands for the Adem relations: Sq Sq =
i j
i [2 ]
j1 Sq i+j + i
k=1
jk1 Sq i+j k Sq k , i 2k
for i, j > 0 with i < 2j . A Batalin-Vilkovisky algebra is a Gerstenhaber algebra equipped with an extra square-zero unary operator of degree +1 satisfying the following quadratic-linear relation: the bracket is the obstruction to being a derivation with respect to the product -, = (- -) ((-) -) (- (-)). So, the operad BV encoding Batalin-Vilkovosky algebras is a quadratic-linear operad. E XERCISE 19. Show that any operad admits a quadratic-linear presentation. D EFINITION (Quadratic-linear-constant operad). A quadratic-linear-constant operad is a quotient operad P (E ; R) := T (E )/(R) generated by a quadratic-linear-constant data (E ; R): R KI E T (E )(2) . E XAMPLES .
25
Let V be symplectic vector space, with symplectic form : V 2 K. Its Weyl algebra is the quadratic-linear-constant algebra dened by W (V, ) := T (V )/(x y y x (x, y )) . In the same way, let V be a quadratic space, with quadratic form q : V K. Its Clifford algebra is the quadratic-linear-constant algebra dened by Cl(V, q ) := T (V )/(x x q (x)) . The nonsymmetric operad uAs admits the following quadratic-linear-constant presentation: uAs =P , ; , , ,
where the arity 0 generator encodes the unit. The properad uF rob encoding unital and countial Frobenius bialgebras admits the following quadratic-linear-constant presentation: uF rob := P , , ; , , , , , , ,
This notion classies 2-dimensional Topological Quantum Field Theories [Koc04]. E XERCISE 20. Let E be an S-module and let R be a sub-S-module of T (E ). We consider the category of operads T (E ) P under T (E ) such that the composite R T (E ) P is zero. A morphism in this / Q such that the following diagram category is a morphism of operads P /P T (E ) ii ii ii " Q is commutative. Show that the quotient operad P (E ; R) = T (E )/(R) is the initial object of this category. 3.4. Koszul duality theory. The Koszul duality theory rst applies to operads having a quadratic presentation P := P (E ; R) = T (E )/(R), R T (E )(2) . Working in the opposite category of vector spaces, that is changing the orientation of every map, one denes the notion of a cooperad. Dually, the quadratic data (E ; R) induces a quadratic cooperad, which, in the Koszul case, provides all the syzygies as follows. D EFINITION (Nonsymmetric Cooperad). A nonsymmetric cooperad is an N-module {Cn }nN equipped with counital and coassociative decomposition maps i1 ,...,ik : Ci1 ++ik Ck Ci1 Cik . The notion of (symmetric) cooperad takes care of the symmetric groups action. Dualizing the partial compositions, one gets partial decompositions (1) : C T (C )(2) , which splits elements of C into two. E XERCISE 21. Show that the linear dual P := C of a nonsymmetric cooperad is a nonsymmetric operad. Give an example of a nonsymmetric operad P whose linear dual P is not a nonsymmetric cooperad with the dual decomposition maps. Give a sufcient nite dimensional condition on P for C := P to be a nonsymmetric cooperad. One can dualize the denition of the free operad to get the notion of the cofree cooperad T c (E ), see [LV10, Section 5.7.7] for more details. It shares the same underlying labeled-tree module as the free operad, with the decomposition maps given by the pruning of trees. (Since the notion of a cooperad is not the exact linear dual of the notion of an operad, as the above exercise shows, this cooperad only satises the universal property of the cofree cooperad only among connected cooperads, that is cooperads for which the iteration
26
BRUNO VALLETTE
of the decomposition maps produces only nite sums, cf. loc. cit.) Let (E, R) be a quadratic data, we consider the category dual to the one introduced in Exercise 20: this is the category of cooperads C T c (E ) over T c (E ) such that the following composite C T c (E ) T c (E )(2) /R is zero. D EFINITION (Quadratic cooperad). The quadratic cooperad C (E ; R) is the terminal object in the aforementioned category. D EFINITION (Koszul dual cooperad). The Koszul dual cooperad is the quadratic cooperad P := C (sE ; s2 R) , where s stands for the degree +1 suspension shift. Theorem 14 (Koszul duality theory [GK94, GJ94]). To any quadratic data (E ; R), there is a morphism of dg operads P := T (s1 P ), d2 P (E ; R) where P := P /I and where d2 is the unique derivation which extends the cooperad structure on P . D EFINITION . A quadratic operad P is called a Koszul operad, when the above map is a quasi-isomorphism. In this Koszul case, all the syzygies and the differential map of the resolution are produced at once by the quadratic data under the universal property of a quadratic cooperad. There are many methods for proving that P is a Koszul operad; we will give an elementary one in the next section. The careful reader might object that it is not a common task to compute a quadratic cooperad, though this is the proper conceptual object which produces all the syzygies. In general, one proceeds as follows. We consider the Koszul dual operad P ! dened as (some desuspension of) the linear dual of the Koszul dual cooperad P . Exercise 21 shows that P is always an operad. Proposition 15 (Koszul dual operad [GK94, GJ94]). When the generating space E is nite dimensional, the Koszul dual operad admits the following quadratic presentation P! = P (E ; R ) , where E = E (2) sgnS2 , when E is concentrated in arity 2 and degree 0. Finally, the method consists in computing the Koszul dual operad, via this formula, and linearizing the result to get the space of syzygies together with the differential. (For more details on how to compute R , we refer the reader to [LV10, Section 7.6].) E XAMPLES . The algebra D! , Koszul dual to the algebra of dual numbers D := T ( )/( 2 ), is the algebra of formal series in one variable D! = K[[ ]]. The coalgebra D , Koszul dual to the algebra of dual numbers D, is the cofree coassociative coalgebra T c (s ) on one generator. The algebra of dual numbers is Koszul. A D -module, also know as multicomplex in the literature, is a chain complex (A, d) equipped with linear maps dn : A A, such that the following identities hold ddn + (1)n dn d =
i+j =n i,j 1 The orthogonal RAs of the associativity relation of the ns operad As is again generated by the associator
for n 1,
|dn | = n 1, for any n 1 .
(1)i di dj ,
27
thanks to the signs and the suspensions. So the ns operad As is Koszul autodual: As! = As. Finally the Koszul resolution A As coincides with the resolution given in Section 3.2. The presentations of the operads Lie and Com, given in 3.3, show that Lie! = Com. The operad Lie is Koszul, see next section. A homotopy Lie algebra, also called an L -algebra, is a dg module (A, d) equipped with a family of skew-symmetric maps n : An A of degree | n | = n 2, for all n 2, which satisfy the relations sgn( )(1)(p1)q ( p 1 q ) = ( n ) ,
p+q =n+1 p,q>1 1 Sh p,q
where Shp,q stands for the set of (p, q )-shufes, that is permutations Sp+q such that (1) < < (p) and (p + 1) < < (p + q ) . E XERCISE 22. Let (A, d, {n }n2 ) be an A -algebra structure on a dg module A. Show that the anti-symmetrized maps n : An A, given by
n
:=
Sn
sgn( )n ,
endow the dg module A with an L -algebra structure. Hint. Use the morphism of cooperads Lie Ass induced by the morphism of operads Lie Ass. The presentations of the operads Com and Lie show that Com! = Lie. The operad Com is Koszul, see next section. A homotopy commutative algebra, also called a C -algebra, is an A -algebra (A, d, {n }n2 ) such that each map n : An A vanishes on the sum of all (p, q )-shufes for p + q = n. This is proved by considering the epimorphism of cooperads Ass Com and by showing that its kernel is given by the sum of the shufes (Rees Theorem [Ree58]). The operad Ger! , Koszul dual to the operad Ger encoding Gerstenhaber algebras, is equal to Ger, up to some suspension. The notion of G -algebra is made explicit in [GCTV11, Section 2.1]. The Koszul duality theory was extended to properads in [Val07]. The properads BiLie and F rob are sent to one another under the Koszul dual functor. In the same way, the properads BiLie and F rob are sent to one another: BiLie F rob
!
&
BiLie F rob .
One can prove, by means of distributive laws, that the properads BiLie and F rob are Koszul. It is still an open conjecture to prove that the properads BiLie and F rob are Koszul. This theory denes the notion of homotopy Lie bialgebra (resp. homotopy involutive Lie bialgebra). One makes it explicit by using that the coproperad structure on F rob (resp. on F rob ) is obtained by dualizing the result of Exercice 18 (resp. Exercice 17) , see [Mer06b] (resp. [DCTT10]) for more details. This theory also denes the notion of homotopy Frobenius bialgebra and homotopy involutive Frobenius bialgebra, whose explicit form is more involved. When the operad P := P (E ; R) admits a quadratic-linear presentation, R E T (E )(2) , we rst consider the homogeneous quadratic operad q P := P (E, qR) , where qR is the image of R under the projection q : T (E ) T (E )(2) . On the Koszul dual cooperad q P , we dene a square-zero coderivation d , which takes care of the inhomogenous relations. The Koszul dual dg cooperad P of P is dene by the dg cooperad (q P , d ). Theorem 16 (Inhomogeneous Koszul duality theory [GCTV11]). For any quadratic-linear data (E ; R), there is a morphism of dg operads P := T (s1 q P ), d1 + d2 P (E ; R)
28
BRUNO VALLETTE
where d1 is the unique derivation which extends d . A quadratic-linear operad P is called a Koszul operad, when the above map is a quasi-isomorphism. This happens when the generating space E is minimal, when the space of relations R is maximal and when the homogeneous quadratic operad q P is Koszul. E XAMPLES . The Koszul dual dga coalgebra of the universal enveloping algebra U (g) is the following dga coalgebra U (g) = (c (sg), dCE ), where c stands for the cofree exterior coalgebra and where dCE is the Chevalley-Eilenberg boundary map dening the homology of the Lie algebra g. The linear dual of the Koszul dual dga coalgebra of the Steenrod algebra A2 is a dga algebra, which is anti-isomorphic to the algebra of [BCK+ 66]. Notice that its homology gives the second page of the Adams spectral sequence which computes homotopy groups of spheres. The examples of the Steenrod algebra and the universal enveloping algebra of a Lie algebra prompted the Koszul duality theory, which was created by Stewart Priddy in [Pri70] for associative algebras. The operad BV is an inhomogeneous Koszul operad, see [GCTV11], with Koszul dual dg cooperad BV = (T c (s) Ger , d ) . The dual t d of the differential is equal to (s) dCE , where dCE is the Chevalley-Eilenberg boundary map dening the homology of the free Lie algebra. The notion of BV -algebra is made explicit in [GCTV11, Section 2.3]. When the operad P := P (E ; R) admits a quadratic-linear-constant presentation R KI E T (E )(2) , we endow the Koszul dual cooperad q P with a coderivation d : q P q P , which takes care of the linear part of the relations, and with a curvature : q P KI, which takes care of the constant part of the relations. All together, these two maps satisfy the same relations as the curvature in geometry. Theorem 17 (Curved Koszul duality theory [HM10]). For any quadratic-linear-constant data (E ; R), there is a morphism of dg operads P := T (s1 q P ), d0 + d1 + d2 P (E ; R) where d0 is the unique derivation which extends the curvature . A quadratic-linear-constant operad P is called a Koszul operad, when the above map is a quasi-isomorphism. This happens under the same conditions as in the quadratic-linear case. Notice that, in this curved case, the Koszul dual cooperad (q P , d , ) is not a dg object, whereas the Koszul resolution P is a dg operad. E XAMPLES . The curved Koszul duality theory was settled for algebra in [Pos93, PP05], where the examples of the Weyl algebras and the Clifford algebras are treated. The nonsymmetric operad uAs is treated in detail in [HM10] and is proved therein to be Koszul. A homotopy unital associative algebra, or uA -algebra, is a dg module (A, d) equipped with a family of maps (n|S |) S S = : A A, | | = n 2 + | S | , n n where the set S runs over the subsets of {1, . . . , n} for any integer n 2, and where S = {1}, for n = 1. We choose to represent them by planar corollas with n leaves and with the inputs in S
29
labelled by a cork. They satisfy the following relations: , ,
This notion of a uA -algebra, obtained by the Koszul duality theory, coincides with the notion of an A -algebra with a homotopy unit given and used in [FOOO09a, FOOO09b]. The associated topological cellular operad was recently constructed by F. Muro and A. Tonks in [MT11]. No Koszul resolution have been proved, so far, for the properad uF rob. However, Joseph Hirsh and Joan Mill` es introduced in [HM10] a curved bar-cobar resolution of it. 3.5. Rewriting method. In this section, we explain the rewriting method, which provides a short algorithmic method, based on the rewriting rules given by the relations, to prove that an operad is Koszul. Let us rst explain the rewriting method for binary quadratic nonsymmetric operads P (E ; R), with the example of the ns operad As in mind. Step 1. We consider an ordered basis 1 < 2 < < k for the generating space E2 of binary operations. E XAMPLE . The ns operad As is generated by one operation of arity 2: .
Step 2. The weight 2 part of the free ns operad T (E ) is spanned by the left combs i 1 j and by the right combs i 2 j . We consider the following total order on this set: for any i, j, i 2 j < i 1 j , i a j < k a l , whenever i < k, a = 1 or 2, and for any j, l, i a j < i a l , whenever j < l, a = 1 or 2. The operad P is determined by the space of relations R, which is spanned by a set of relators written in this basis as r = i a j i,a,j , i,a,j k,b,l k b l , k,b,l K and = 0 , where the sum runs over the indices satisfying i a j > k b l . The operation i a j is called the leading term of the relator r. One can always suppose that is equal to 1, that the leading terms of the set of relators are all distinct and that there is no leading term of any other relator in the sum on the right hand side. This is called a normalized form of the presentation. E XAMPLE . There is only one relator in As: r= .
Step 3. Every relator gives rise to a rewriting rule in the operad P : i a j E XAMPLE . i,a,j k,b,l k b l .
30
BRUNO VALLETTE
Given three generating binary operations i , j , k , one can compose them in 5 different ways: they correspond to the 5 planar binary trees with 3 vertices. Such a monomial, i.e. decorated planar tree, is called critical if the two sub-trees with 2 vertices are leading terms. E XAMPLE . For the nonsymmetric operad As, the left comb is the only following critical monomial.
Step 4. There are at least two ways of rewriting a critical monomial ad libitum, that is, until no rewriting rule can be applied any more. If all these ways lead to the same element, then the critical monomial is said to be conuent. E XAMPLE . The critical monomial of As gives rise to the conuent graph.
Conclusion. If every critical monomial is conuent, then the ns operad P is Koszul.
E XERCISE 23. Consider the same example but with the modied associativity relation =2 .
Show that the only critical monomial is not conuent. (It can be proved that this ns operad is not Koszul.) The general method holds for quadratic operads P (E ; R) as follows. We forget the symmetric group action and we consider a K-linear ordered basis 1 < 2 < < k for the generating space E . Then, the only difference lies in Step 2, where one has to use a good basis for the free operad T (E ). To this end, we introduce the set T of shufe trees , which are planar rooted trees equipped with a bijective labeling of the leaves by integers {1, 2, . . . , n} satisfying the following condition. First, we label each edge by the minimum of the labels of the inputs of the above vertex. Secondly, we require that, for each vertex, the labels of the inputs, read from left to right, are increasing.
31
Proposition 18 (Shufe tree basis [Hof10, DK10]). The set T ({1, 2, . . . , k }) of shufe trees with vertices labelled by 1, 2, . . . , k forms a K-linear basis of the free operad T (E ). We then consider a suitable order on shufe trees, that is a total order, which makes the partial compositions into strictly increasing maps. For instance, the path-lexicographic order, dened in [Hof10], is a suitable order. On binary shufe trees with 2 vertices, it is given by 3 3 2 b 2 b b b b b b b 0123 7654 7654 0123 1` 1` 2 1 < ` ` ` ` ` ` 7654 0123 7654 0123 1 1 2 b 3 3 2 b b b b b b b 0123 7654 89:; ?>=< 1 k < < 1 ` < 1` < ` ` ` ` ` ` 0123 7654 0123 7654 2 1
2c 3 1 a 1 a 1 a 3 2 3 c aa aa aa c c 89:; ?>=< 7654 0123 0123 7654 0123 7654 2 < 3 < 2 1a 1a 2a k < 1 ` < aa aa aa ` < . ` ` a a a 7654 0123 7654 0123 7654 0123 89:; ?>=< 1 1 1 k
The rewriting method relies on the following result. Theorem 19 (Rewriting method). Let P (E ; R) be a quadratic (resp. nonsymmetric) operad. If its generating space E admits an ordered basis, for which there exists a suitable order on shufe (resp. planar) trees such that every critical monomial is conuent, then the (resp. nonsymmetric) operad P is Koszul. Proof. This theorem is proved by a version of the Diamond Lemma of George M. Bergman [Ber78] for operads, see [DK10] and [LV10, Chapter 8].
E XERCISE 24. Apply the rewriting method to the operad Lie to show that it is Koszul. First, prove that, under the path-lexicographic order, the Jacobi relation becomes the following rewriting rule:
Finally, show that there is only one critical monomial, which is conuent:
32
BRUNO VALLETTE
R EMARK . Notice that the graph given here is a compact version of the full rewriting graph of the operad Lie. There is another way to draw this rewriting diagram, which gives the Zamolodchikov tetrahedron equation. It leads to the categorical notion of Lie 2-algebra, see Baez-Crans [BC04, Section 4]. E XERCISE 25. Apply the rewriting method to the operad Com, to show that it is Koszul. When the rewriting method applies, the operad P admits a K-linear basis made up of some planar trees called a Poincar e-Birkhoff-Witt basis, PBW basis for short. The PBW basis is given by the set of shufes (resp. planar) trees with vertices labelled by 1, 2, . . . , k such that no sub-tree with 2 vertices is a leading term. E XAMPLES . The ns operad As admits a PBW basis made up of the right combs:
It is interesting to notice that one recovers exactly Mac Lanes coherence theorem for (non-unital) monoidal categories [ML98]. This is not a so big surprise since both actually relies on the Diamond Lemma. With this remark in mind, the reading of [ML95, Section VII-2] enjoys another savor. The M Rn -trees, dened in Section 2.4, form the PBW basis of the operad Lie. It appears quite often in the literature when one wants to prove that some representation of the symmetric group, coming either from algebraic topology or algebraic combinatorics, is isomorphic to Lie(n), see for instance [Coh76, RW02, Tou06, ST09]. The notion of operadic PBW basis provides a special basis for quotient operads P (E ; R) = T (E )/(R). It was introduced by Eric Hoffbeck in [Hof10]. The dual notion of Gr obner basis for the operadic ideal (R) was introduced by Vladimir Dotsenko and Anton Khoroshkin in [DK10].
33
4. H OMOTOPY T RANSFER T HEOREM The purpose of this section is to extend all the results given in Section 1 on the level of associative algebras to any category of algebras over a Koszul operad P . We rst state a Rosetta stone, which gives four equivalent denitions of a P -algebra. The Homotopy Transfer Theorem can be proved with the third equivalent denition and the notion of an -morphism is dened with the fourth one. This provides us with all the necessary tools to study the homotopy theory of P -algebras. 4.1. The Rosetta stone for homotopy algebras. In this section, we give three other equivalent denitions of a P -algebra. Let C be a dg cooperad and let Q be a dg operad. Their partial compositions and decompositions endow the space of S-equivariant maps HomS (C , Q) :=
nN
HomSn (C (n), Q(n))
with a dg Lie algebra structure HomS (C , Q), [ , ], , called the convolution algebra. D EFINITION (Twisting morphism). A twisting morphism is a solution : C Q of degree 1 to the the Maurer-Cartan equation 1 () + [, ] = 0 2 in the convolution dg Lie algebra. The associated set is denoted by Tw(C , Q). This notion is sometimes called twisting cochain in algebraic topology. This bifunctor can be represented both on the left-hand side and on the right-hand side by the following bar-cobar adjunction. : (augmented) dg cooperads (conilpotent) dg operads : B .
D EFINITION (Bar and cobar constructions). The bar construction BQ of an (augmented) dg operad (Q, dQ ) is the dg cooperad BQ := (T c (sQ), d1 + d2 ) , where d1 is the unique coderivation which extends dQ and where d2 is the unique coderivation which extends the partial compositions of the operad Q. (An augmented operad is an operad such that Q = IQ is a morphism of operads). The cobar construction C of a (coaugmented) dg cooperad (C , dC ) is the dg operad C := (T (s1 C ), d1 + d2 ) , where d1 is the unique derivation which extends dC and where d2 is the unique derivation which extends the partial decompositions of the cooperad C . Notice that the Koszul resolution of Section 3.4 is given by the cobar construction of the Koszul dual cooperad: P = P . We apply the preceding denitions to the dg cooperad C := P and to the dg operad Q := EndA . Let us begin with HomS (P , EndA ) =
nN
HomSn (P (n), Hom(An , A)) = Hom

nN
P (n) Sn An , A
n We denote by P (A) := ; this is the cofree P -coalgebra structure on A. Since nN P (n) Sn A coderivations on cofree P -coalgebras are completely characterized by their projection P (A) A onto the space of generators, we get HomS (P , EndA ) = Coder(P (A)). Under this isomorphism, twisting morphisms correspond to square-zero coderivations, that we call codifferentials and the set of which we denote Codi(P (A)).
34
BRUNO VALLETTE
Theorem 20 (Rosetta Stone). The set of P -algebra structures on a dg module A is equivalently given by Homdg Op ( P , EndA ) = Tw(P , EndA ) = Codi(P (A)) . = Homdg Coop (P , B EndA )
Denition Deformation theory HTT morphism
We indicated which equivalent denition suits which question best. The rst one was used in Section 3 to dene P -algebras. The second one is used to study the deformation theory of P -algebras and P algebras, see [LV10, Section 12.2]. The third one will be used in Section 4.3 to state the Homotopy Transfer Theorem (HTT). We are going to use the fourth one in the next section to dene the notion of an -morphism between P -algebras. E XERCISE 26. Describe the aforementioned four equivalent denitions in the case of A -algebras. Hint. The Koszul resolution As was described in 3.2. The convolution dg Lie algebra Tw(As , EndA ) is the Hochschild cochain complex together with the Gerstenhaber bracket [Ger63]. Finally, the data of an A -algebra structure on A is equivalently given by a square-zero coderivation on the noncounital cofree c coassocative coalgebra T (sA). E XERCISE 27. Describe the aforementioned four equivalent denitions in the case of L -algebras. Hint. The Koszul resolution Lie was described in 3.4. The convolution dg Lie algebra Tw(Lie , EndA ) is the Chevalley-Eilenberg cochain complex together with the Nijenhuis-Richardson bracket [NR66, NR67]. Finally, the data of an L -algebra structure on A is equivalently given by a square-zero coderivation on c the noncounital cofree cocommutative coalgebra S (sA). 4.2. -morphism. We use the fourth denition given in the Rosetta stone to dene a notion of morphisms between P -algebras with nice properties. D EFINITION (-morphism). Let A and B be two P -algebras. An -morphism A B is a morphism P (A) P (B ) of dg P -coalgebras. B between A and
Two such morphisms are obviously composable and the associated category is denoted -P -alg. E XERCISE 28. Show that, in the case of the ns operad As, ones recovers the denition of an A -morphism given in Section 1.6. E XERCISE 29. Make explicit the notion of -morphism between L -algebras. This notion is called L -morphism in the litterature. As a morphism to a cofree P -coalgebra, an -morphism is completely charactrized by its projection P (A) B onto the space of generators. Such a data is equivalent to giving a morphism of S-modules A n P EndA , B )}nN . The commutativity of the respective codifferentials B , where EndB := {Hom(A translates into some relation satised by this latter map P EndA B. E XERCISE 30. Make this relation explicit in terms of the decomposition maps and the partial decompositions of the cooperad P . For any -morphism f : A B , the image of I P produces a chain map f1 : A B . D EFINITION (-isomorphism and -quasi-isomorphism). When the map f1 is an isomorphism (resp. a quasi-isomorphism), the -morphism f is called an -isomorphism (resp. an -quasi-isomorphism). Proposition 21. The -isomorphisms are the invertible morphisms of the category -P alg. Proof. We use the weight grading on the Koszul dual cooperad P T c (sE ) and we notice then that the proof is similar to the proof that power series a1 x + a2 x2 + with invertible rst term are invertible. 4.3. Homotopy Transfer Theorem. We now have all the tools to prove the following result, with explicit constructions. Theorem 22 (Homotopy Transfer Theorem [GCTV11]). Let P be a Koszul operad and let (H, dH ) be a homotopy retract of (A, dA ):
h
(A, dA ) o
p i
/ (H, d ) H
IdA ip = dA h + hdA and i quasi-isomorphism.
35
Any P -algebra structure on A can be transferred into a P -algebra structure on H such that i extends to an -quasi-isomorphism. R EMARK . The existence part of the theorem can also be proved by model category arguments, see Clemens Berger and Ieke Moerdijk [BM03] and Benoit Fresse [Fre09]. Proof. The proof is based on the third denition Homdg Coop (P , B EndA ) of the Rosetta stone together with a morphism of dg cooperads : B EndA B EndH , that we describe below:
Homdg Coop (P , B EndA )
Homdg Coop (P , B EndH ) Transferred structure := .
Initial structure
Lemma 23. [VdL03, Theorem 5.2] Let (H, dH ) be a homotopy retract of (A, dA ). The unique morphism of cooperads which extends : T c (sEndA ) EndH :
is a morphism of dg cooperads : B EndA B EndH between the cobar constructions. Explicitly, under the identication Tw(P , EndA ) = Homdg Coop (P , B EndA ), if : P EndA denotes the initial P -algebra on A, then the transferred P -algebra : P EndH on H is equal to the following composite P
P T c (P ) (type of algebra)
(initial structure)
T c (s )
T c (sEndA )
(homotopy data)
EndH ,
where the rst map P : P T c (P ) is given by all the possible iterations of the partial decomposition maps of the cooperad P . This composite is made up of 3 independent terms corresponding respectively to the type of algebraic structure considered, to the initial algebra structure, and to the homotopy data. E XAMPLES . In the example of the ns operad As, its Koszul dual cooperad is the linear dual As , up to suspension. So, the full decomposition map As : As T c (As ) produces all the planar trees out of any corolla:
36
BRUNO VALLETTE
The second map labels their vertices with the initial operations {n }n2 according to the arity. The third map labels the inputs by i, the internal edges by h, and the root by p. Finally, we get the formula given in Theorem 4. The Koszul dual cooperad Lie = Com of the operad Lie is the linear dual of Com, up to suspension. So, the full decomposition map Com : Com T c (Com ) splits any corolla into the sum of all rooted trees. If { n : An A}n2 stands for the initial L -algebra structure on A, then the transferred L -algebra structure {ln : H n H }n2 is equal to ln =
tRTn
p t( , h) in ,
where the sum runs over rooted trees t with n leaves and where the notation t( , h) stands for the n-multilinear operation on A dened by the composition scheme t with vertices labeled by the operations k and the internal edges labeled by h. Recall that a C -algebra is an A -algebra whose structure maps n : An A vanish on the sum of all (p, q )-shufes for p + q = n. Using the morphism of cooperads Ass Com and the aforementioned boxed formula, we leave it to the reader to prove that the planar tree formula for the transfer of A -algebra structures applies to C -algebras as well. See [CG08] for a proof by direct computations. The data of a dg D-module is equivalent to the data of a bicomplex (A, d, ), that is a module equipped with two anti-commuting square-zero unary operators. Considering the homotopy retract (H, dH ) := (H (A, d), 0) of (A, d), the transferred D -module structure {dn : H H }n1 is a lifted version of the spectral sequence associated to the bicomplex (A, d, ). Indeed, one can easily see that the formula for the transferred structure dn := p h h h i
n times
corresponds to the diagram-chasing formula [BT82, 14] giving the higher differentials of the spectral sequence. Such a formula goes back it least to [Shi62]. The great advantage with this conceptual presentation is that one can prove the HTT for Koszul properads as well, using the very same arguments. The bar and cobar constructions were extended to the level of properads in [Val07]. The only missing piece is a Van der Laan type morphism for graphs with genus instead of trees. Such a morphism is given in [Val11b]. This settles the HTT for homotopy (involutive) Lie bialgebras and for homotopy (involutive) Frobenius bialgebras, for instance. The quasi-ismorphism i : H A extends to an -quasi-isomorphism : H A dened by the same formula as the one giving the transferred structure but replacing the map p labeling the root by the homotopy h. R EMARK . The Homotopy Transfer Theorem should not be confused with the Homological Perturbation Lemma. One can prove the HTT with it, see [Ber09]. The other way round, the HTT applied to the algebra of dual numbers gives the perturbation lemma. 4.4. Homotopy theory of P -algebras. Theorem 24 (Fundamental theorem of -quasi-isomorphisms [LV10]). If there exists an -quasi-iso morphism A B between two P -algebras, then there exists an -quasi-isomorphism in the opposite = direction B A, which is the inverse of H (A) H (B ) on the level on holomogy. This is the main property of -quasi-isomorphisms, which does not hold for quasi-isomorphisms of dg P -algebras. These two notions of quasi-isomorphisms are related by the following property: there exists a zig-zag of quasi-isomorphisms of dg P -algebras if and only there exists a direct -quasi-isomorphism A B

B.
D EFINITION (Homotopy equivalence). Two dg P -algebras (resp. two P -algebras) are homotopy equivalent if they are related by a zig-zag of quasi-isomorphisms of dg P -algebras (resp. by an -quasiisomorphism).
37
The underlying homology groups H (A) of any dg P -algebra A carry a natural P -algebra structure. Moreover, the Homotopy Transfer Theorem 22 allows us to endow H (A) with a P -algebra structure, with trivial differential, and which extends this induced P -algebra structure. D EFINITION (Operadic Massey products). We call operadic Massey products the operations making up this transferred P -algebra structure on the homology groups H (A). D EFINITION (Formality). A dg P -algebra (A, d) is called formal if it is homotopy equivalent to the P algebra H (A) equipped with the induced structure: (A, d) (H (A), 0) . Proposition 25. If the higher operadic Massey products vanish, then the dg P -algebra is formal. Proof. The proof is a corollary of the HTT, Theorem 22. In other words, when the higher operadic Massey products vanish, the homology P -algebra H (A) has the same homotopy type as the initial dg P -algebra A. To study the general case, we will need the following result. Proposition 26 (Rectication property [DTT07]). Any P -algebra A admits a dg P -algebra, denoted Rect(A), which is -quasi-isomorphic to it A

Rect(A) .
The rectication property together with the HTT provide us with two -quasi-isomorphisms
A o /o /o /o H (A) /o o/ o/ / Rect(A) ,
where H (A) is considered as a P -algebra equipped with the operadic Massey products. Therefore, the data of the operadic Massey products on the homology groups allows us to recover the homotopy type of the initial dg P -algebra. So the Massey products faithfully encode the homotopy type of A. In the homotopy class of any P -algebra (A, d), there is a P -algebra H (A) with trivial differential, by the HTT, and there is a strict dg P -algebra Rect(A), by the Rectication property. In the rst case, the underlying space is rather small but the algebraic structure is more complex. In the latter case, the underlying space is pretty big (one has to fatten A to get Rect(A)) but the algebraic structure is simpler. We informally call this phenomenon the Heisenberg uncertainty principle of homotopical algebra: in general, homotopy classes of P -algebras cannot be represented by two small data, space and structure, at the same time. One can dene a suitable notion of -homotopy, denoted h , between -morphisms, see [Val11a], which allows us to state the following result. Theorem 27 (Homotopy category [Val11a]). The following categories are equivalent Ho(dg P -alg) = P alg/ h = -dg P -alg/ h . Proof. The rst equivalence is proved by endowing the category of dg P -coalgebras with a model category structure. The last equivalence with the category of dg P -algebras equipped with the -morphisms is given by the Rectication property. 4.5. Applications. In this section, we conclude with a non-exhaustive, but not so short, list of elds where the aforementioned operadic homotopical algebra is used. A LGEBRA . Chevalley-Eilenberg cohomology of Lie algebras. The cohomology groups of a Lie algebra g, with coefcients in S (g) and the adjoint representation, carry a homotopy BV -algebra structure [DCV11]. The cohomology of the Lie algebra L1 of polynomial vector elds over the line K1 is a C -algebra generated 1 by HCE (L1 ) [Mil10].
38
BRUNO VALLETTE
Bar construction. The bar construction for A -algebras carries a natural commutative BV -algebra structure [TTW11]. Vertex algebras. The study of vertex algebras [Bor86] yields to BV-algebras. Lian and Zuckerman [LZ93] conjectured that their structure of BV-algebra on the homology groups of a Topological Vertex Operator Algebra lifts to a homotopy BV-algebra structure on the TVOA itself. This was proved, with explicit formulae, in [GCTV11]. (Cyclic) Deligne conjecture. The (cyclic) Deligne conjecture states that the (BV) Gersenhaber algebra structure on the Hochschild cohomology groups of an (cyclic unital) associative algebra can be lifted to a homotopy (BV) Gersenhaber algebra structure on the cochain level. This conjecture received several proofs in [Tam98, Vor00, KS00, MS02, BF04, Kau08, Kau07, TZ06, Cos07, KS09]. D EFORMATION THEORY, Q UANTUM ALGEBRA . Deformation-Quantization of Poisson manifolds. The rst proof of the deformation-quantization of Poisson manifolds, given by Maxim Kontsevich in [Kon03, Wei95], relies on an explicit L -quasi-isomorphism to prove a formality result. Drinfeld associators, Grothendieck-Teichm uller group. The second proof of the deformation-quantization of Poisson manifolds, given by Dimitry Tamarkin in [Tam98], relies on the homotopy properties of G algebras. It relies on the obstruction theory; so it is less explicit than Kontsevichs proof. It however unveils the links with the Drinfeld associators and the GT groups. A LGEBRAIC T OPOLOGY. Massey products. The Massey products on the cohomology groups of a topological space are related to the operad A , see Section 1.4. Homotopy groups of spheres. The second page of the Adams spectral sequence which computes the homotopy groups of spheres is equal to the homology of the algebra, which is the Koszul dual dg algebra of the Steenrod algebra, see [BCK+ 66]. Spectral Sequences. We have seen in Section 4.3 that spectral sequences actually come from the HTT applied to the algebra of dual numbers. The advantage of this point of view on bicomplexes versus spectral sequences, is that the HTT can be applied to bicomplexes equipped with a deformation retract whose boundary map is not necessarily trivial. For instance, the cyclic bicomplex of a unital associative algebra, which involves the boundary maps b, b and the cyclic operator, cf. [LQ84, Lod98], admits a deformation retract made up of the the columns involving only b. Applying the HTT to it gives automatically Connes boundary map B . So we recover the fact that, in cyclic homology theory, the (b, B )-bicomplex is quasiisomorphic to the cyclic bicomplex. Iteration of the bar construction. The method proposed by Benoit Fresse in [Fre10] to iterate the bar construction for E -algebras relies on the HTT for a cobrant E -operad. (An E -algebra is a homotopy version of a commuative algebra, where both the associativity relation and the symmetry property are relaxed up to higher homotopies). Rational Homotopy Theory. Minimal models in Dennis Sullivans approach to Rational Homotopy Theory [Sul77] are quasi-free commutative algebras on the rational homotopy groups of topological spaces endowed with an L -(co)algebra structure. In Dan Quillens approach to RHT [Qui69], the minimal models are quasi-free Lie algebras generated by the cohomology groups of the space endowed with a C -(co)algebra structure. String Topology. String Topology is full of higher structures, to name but a few: Batalin-Vilkovisky algebra in loop homology [CS99, CV06] and (homotopy) involutive Lie bialgebra in (equivariant) string homology [CS04]. D IFFERENTIAL G EOMETRY . Lie algebroids. The structure of Courant algebroids is shown to induce a structure of homotopy Lie algebra by Dmitry Roytenberg and Alan Weinstein in [RW98].
39
K ahler manifolds. The dg commutative algebra of differential forms of a K ahler manifold is shown to be formal in [DGMS75]. It was also proved in loc. cit. that this is equivalent to the uniform vanishing of the higher Massey products. Lagrangian submanifolds, Floer Homology, Symplectic Field Theory. Structures of A -algebras [Fuk02], uA -algebras [FOOO09a, FOOO09b] and BiLie -algebras [EGH00, CFL] play a crucial role in these elds. F-manifold, Nijenhuis structure and Poisson manifold. Sergei Merkulov developed in [Mer04, Mer05, Mer06a] a programme called operadic proles, which establishes a new link between differential geometry and higher algebra. He thereby describes several differential geometric structures (Hertling-Manin, Nijenhuis, and Poisson) in terms of homotopy algebras, i.e. algebras over operadic Koszul resolutions. Poincar e duality. The study of the Poincar e duality of oriented closed manifolds is related to the homotopy (co)unital Frobenius bialgebra structure [Wil07, HM10, Mil11] on the differential forms of the manifold. Fluid mechanics. Dennis Sullivan proposed in [Sul10] a programme to solve the Euler and NavierStockes equations, which describe uid motion, using the HTT for BV -algebras. A LGEBRAIC G EOMETRY . Moduli spaces of algebraic curves Mg,n and Mg,n , Gromov-Witten invariants. The two homology operads H (M0,n+1 ) and H (M0,n+1 ) are Koszul dual to each other, see [Get95]. Frobenius manifolds, Quantum cohomology. The space H (M0,n+1 ) provides the generators of the minimal model (quasi-free resolution without internal differential) of the operad BV , see [DCV11]. This allows us to dene the notion of a homotopy Frobenius manifold. Using the HHT for Lie algebras, S. Barannikov and M. Kontsevich [BK98], and Y.I. Manin [Man99] proved that the underlying homology groups of some dg BV-algebras carry a Frobenius manifold structure, i.e. an algebra over H (M0,n+1 ). Using the aforementioned minimal model for BV and the HTT for homotopy BV algebras, we endowed in [DCV11] the homology groups with a homotopy Frobenius manifold structure, i.e. an algebra parametrized by H (M0,n+1 ), extending the Barannikov-Kontsevich-Manin structure, cf. loc. cit. This latter structure keeps faithfully track of the homotopy type of the initial dg BV -algebra. Mirror symmetry conjecture. The Mirror Symmetry conjecture [Kon95] claims that the Fukaya category of Lagrangian submanifolds of a Calabi-Yau manifold M (A-side) should be equivalent to the bounded derived category of coherent sheaves on a dual Calabi-Yau manifold M (B-side). The Fukaya category is actually an A -category. The relevant algebraic structure on the B-side is the homotopy Frobenius manifold stated in the previous point, where one starts for the dg BV-algebra of the Dolbeault cochain complex of the Calabi-Yau manifold M. M ATHEMATICAL P HYSICS . Feynman diagrams, Batalin-Vilkovisky formalism, Renormalization theory. One of the highlights of this operadic homotopy theory lies in the following result by Sergey Merkulov [Mer10]. He proved that the Batalin-Vilkovisky formalism is equivalent to the HTT for unimodular Lie bialgebras. In other words, the classical Feynman diagrams are exactly the graphs appearing in the HTT formula for the wheeled properad encoding unimodular Lie bialgebras. BRST cohomology. The BRST cohomology carries a BV -algebra structure [GCTV11]. Field theories. The various types of elds theories (Topological Quantum Field Theory [Ati88], Conformal Field Theory [Seg04], String Field Theories [WZ92, Zwi93], etc.) yield to various types of homotopy structures ( L [FLS02], uF rob [HM10], OCHA [KS06], BiLie [MS11], etc.). C OMPUTER S CIENCE . Rewriting system. The rewriting method of Section 3.5 is strongly related to the notion of Rewriting system, see [GM09, GM11] for instance.
40
BRUNO VALLETTE
C ONCLUSION Open questions. Here are a few open and interesting questions in this eld of research. (1) Are the Koszul dual properads F rob and BiLie , encoding respectively Frobenius bialgebras and involutive Lie bialgebras, Koszul ? (2) There 4 functors making up the following commutative diagram modular operads cyclic operads / properads / operads .
It was shown in [Get95] that the two graded operads H (M0,n ) and H (M0,n ) are Koszul dual operads and Koszul operads. We conjecture that the two full genera properads associated to the modular operads H (Mg,n+1 ) and H (Mg,n+1 ) are Koszul dual properads and Koszul properads. (3) Develop the Koszul duality for operads in characteristic p. Further reading. At this point, the interested reader is invited to pursue its journey in operad theory and homotopical algebra, with the book [LV10]. The various answers to the exercises proposed here can be found in this reference. Acknowledgements. This survey follows several talks given at the colloquia or seminars of the universities of Lyon, Dublin, Stony Brooke, G ottingen, Glasgow, Paris 6, at the conference Higher Structures 2009 in Z urich, and at the Hayashibara forum on Symplectric Geometry, Non-commutative geometry and Mathematical Physics at MSRI. The author would like to thank the audience, and especially Dennis Sullivan, for their questions and comments, which helped to improve the present text. Many thanks to Olivia Bellier and Jim Stasheff for their corrections and interesting remarks on the rst draft of the present paper. Last but not least, I would like to express my deep gratitude to the Max-Planck Institute f ur Mathematik in Bonn for the long term invitation and for the excellent working conditions. R EFERENCES
Michael Atiyah, Topological quantum eld theories, Inst. Hautes Etudes Sci. Publ. Math. (1988), no. 68, 175186. 18, 39 [BC04] John C. Baez and Alissa S. Crans, Higher-dimensional algebra. VI. Lie 2-algebras, Theory Appl. Categ. 12 (2004), 492538 (electronic). 32 [BCK+ 66] A. K. Bouseld, E. B. Curtis, D. M. Kan, D. G. Quillen, D. L. Rector, and J. W. Schlesinger, The mod p lower central series and the Adams spectral sequence, Topology 5 (1966), 331342. 28, 38 [Ber78] George M. Bergman, The diamond lemma for ring theory, Adv. in Math. 29 (1978), no. 2, 178218. 31 [Ber09] A. Berglund, Homological perturbation theory for algebras over operads, arXiv:0909.3485 (2009). 36 [BF04] Clemens Berger and Benoit Fresse, Combinatorial operad actions on cochains, Math. Proc. Cambridge Philos. Soc. 137 (2004), no. 1, 135174. 38 [BK98] Sergey Barannikov and Maxim Kontsevich, Frobenius manifolds and formality of Lie algebras of polyvector elds, Internat. Math. Res. Notices (1998), no. 4, 201215. 39 [BM03] C. Berger and I. Moerdijk, Axiomatic homotopy theory for operads, Comment. Math. Helv. 78 (2003), no. 4, 805831. 22, 35 [Bor86] Richard E. Borcherds, Vertex algebras, Kac-Moody algebras, and the Monster, Proc. Nat. Acad. Sci. U.S.A. 83 (1986), no. 10, 30683071. 38 [BT82] Raoul Bott and Loring W. Tu, Differential forms in algebraic topology, Graduate Texts in Mathematics, vol. 82, Springer-Verlag, New York, 1982. 36 [BV73] J. M. Boardman and R. M. Vogt, Homotopy invariant algebraic structures on topological spaces, Lecture Notes in Mathematics, Vol. 347, Springer-Verlag, Berlin, 1973. 14 [Car55] Henri Cartan, Sur lit eration des op erations de Steenrod, Comment. Math. Helv. 29 (1955), 4058. 9 [CFL] Kai Cieliebak, Kenji Fukaya, and Janko Latschev, Homological algebra related to surfaces with boundaries, unpublished. 39 [CG08] Xue Zhi Cheng and Ezra Getzler, Transferring homotopy commutative algebraic structures, J. Pure Appl. Algebra 212 (2008), no. 11, 25352542. 36 [Coh76] Frederick R. Cohen, The homology of Cn+1 -spaces, The homology of iterated loop spaces (Cohen, Frederick R. and Lada, Thomas J. and May, J. Peter), Lecture Notes in Mathematics, Vol. 533, Springer-Verlag, Berlin, 1976, pp. vii+490. 15, 32 [Ati88]
41
[Cos07] [CR06] [CS99] [CS04] [CV06] [DCTT10] [DCV11] [DGMS75] [DK10] [DM69] [Dri86] [DTT07] [EGH00] [Eis05] [FLS02] [FOOO09a]
[FOOO09b] [Fre09]
[Fre10] [Fuk02] [GCTV11] [Ger63] [Get95] [GJ94] [GK94] [GM09] [GM11] [Hat02] [Hin97] [HM10] [Hof10] [HT92] [Jon99] [Jon10] [Kad80]
Kevin Costello, Topological conformal eld theories and Calabi-Yau categories, Adv. Math. 210 (2007), no. 1, 165214. 38 Charles W. Curtis and Irving Reiner, Representation theory of nite groups and associative algebras, AMS Chelsea Publishing, Providence, RI, 2006, Reprint of the 1962 original. 16 M. Chas and D. Sullivan, String Topology, ArXiv Mathematics e-prints (1999). 24, 38 Moira Chas and Dennis Sullivan, Closed string operators in topology leading to Lie bialgebras and higher string algebra, The legacy of Niels Henrik Abel, Springer, Berlin, 2004, pp. 771784. 38 Ralph L. Cohen and Alexander A. Voronov, Notes on string topology, String topology and cyclic homology, Adv. Courses Math. CRM Barcelona, Birkh auser, Basel, 2006, pp. 195. 38 Gabriel C. Drummond-Cole, John Terilla, and Thomas Tradler, Algebras over (coFrob), J. Homotopy Relat. Struct. 5 (2010), no. 1, 1536. 27 Gabriel Drummond-Cole and Bruno Vallette, The minimal model for the batalin-vilkovisky operad, Preprint (2011). 37, 39 Pierre Deligne, Phillip Grifths, John Morgan, and Dennis Sullivan, Real homotopy theory of K ahler manifolds, Invent. Math. 29 (1975), no. 3, 245274. 7, 39 Vladimir Dotsenko and Anton Khoroshkin, Gr obner bases for operads, Duke Math. J. 153 (2010), no. 2, 363396. 31, 32 P. Deligne and D. Mumford, The irreducibility of the space of curves of given genus, Inst. Hautes Etudes Sci. Publ. Math. (1969), no. 36, 75109. 16 V. G. Drinfel d, Quantum groups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, Calif., 1986), 1986, pp. 798820. 23 Vasiliy Dolgushev, Dmitry Tamarkin, and Boris Tsygan, The homotopy Gerstenhaber algebra of Hochschild cochains of a regular algebra is formal, J. Noncommut. Geom. 1 (2007), no. 1, 125. 37 Y. Eliashberg, A. Givental, and H. Hofer, Introduction to symplectic eld theory, Geom. Funct. Anal. (2000), no. Special Volume, Part II, 560673, GAFA 2000 (Tel Aviv, 1999). 39 David Eisenbud, The geometry of syzygies, Graduate Texts in Mathematics, vol. 229, Springer-Verlag, New York, 2005, A second course in commutative algebra and algebraic geometry. 22 Ron Fulp, Tom Lada, and Jim Stasheff, sh-Lie algebras induced by gauge transformations, Comm. Math. Phys. 231 (2002), no. 1, 2543. 39 Kenji Fukaya, Yong-Geun Oh, Hiroshi Ohta, and Kaoru Ono, Lagrangian intersection Floer theory: anomaly and obstruction. Part I, AMS/IP Studies in Advanced Mathematics, vol. 46, American Mathematical Society, Providence, RI, 2009. 29, 39 , Lagrangian intersection Floer theory: anomaly and obstruction. Part II, AMS/IP Studies in Advanced Mathematics, vol. 46, American Mathematical Society, Providence, RI, 2009. 29, 39 Benoit Fresse, Operadic cobar constructions, cylinder objects and homotopy morphisms of algebras over operads, Alpine perspectives on algebraic topology, Contemp. Math., vol. 504, Amer. Math. Soc., Providence, RI, 2009, pp. 125 188. 35 , The bar complex of an E -innity algebra, Adv. Math. 223 (2010), no. 6, 20492096. 38 Kenji Fukaya, Floer homology and mirror symmetry. II, Minimal surfaces, geometric analysis and symplectic geometry (Baltimore, MD, 1999), Adv. Stud. Pure Math., vol. 34, Math. Soc. Japan, Tokyo, 2002, pp. 31127. 39 I. Galvez-Carrillo, A. Tonks, and B. Vallette, Homotopy Batalin-Vilkovisky algebras, to appear in Journal of Noncommutative Geometry (2011), arXiv:0907.2246. 27, 28, 34, 38, 39 M. Gerstenhaber, The cohomology structure of an associative ring, Ann. of Math. (2) 78 (1963), 267288. 34 E. Getzler, Operads and moduli spaces of genus 0 Riemann surfaces, The moduli space of curves (Texel Island, 1994), Progr. Math., vol. 129, Birkh auser Boston, Boston, MA, 1995, pp. 199230. 39, 40 E. Getzler and J. D. S. Jones, Operads, homotopy algebra and iterated integrals for double loop spaces, hep-th/9403055 (1994). 26 V. Ginzburg and M. Kapranov, Koszul duality for operads, Duke Math. J. 76 (1994), no. 1, 203272. 26 Yves Guiraud and Philippe Malbos, Higher-dimensional categories with nite derivation type, Theory Appl. Categ. 22 (2009), No. 18, 420478. 39 Y. Guiraud and P. Malbos, Identities among relations for higher-dimensional rewriting systems, Sminaires et Congrs 26 (2011), 145161. 39 Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002. 5 Vladimir Hinich, Homological algebra of homotopy algebras, Comm. Algebra 25 (1997), no. 10. 22 Joseph Hirsh and Joan Mill` es, Curved Koszul duality theory, arXiv.org:1008.5368 (2010). 28, 29, 39 Eric Hoffbeck, A Poincar e-Birkhoff-Witt criterion for Koszul operads, Manuscripta Math. 131 (2010), no. 1-2, 87110. 31, 32 Marc Henneaux and Claudio Teitelboim, Quantization of gauge systems, Princeton University Press, Princeton, NJ, 1992. 22 Vaughan F.R. Jones, Planar algebras, i, ArXiv e-prints (1999), arXiv:math/9909027. 15 , Quadratic tangles in planar algebras, ArXiv e-prints (2010), arXiv:0907.2246. 15 T. V. Kadei svili, On the theory of homology of ber spaces, Uspekhi Mat. Nauk 35 (1980), no. 3(213), 183188, International Topology Conference (Moscow State Univ., Moscow, 1979). 4
42
BRUNO VALLETTE
[Kad82] [Kau07] [Kau08] [KM94] [Koc04] [Kon95] [Kon03] [Kos50] [KS00] [KS01] [KS06] [KS09] [Law63] [LH03] [Lod98]
[LQ84] [LV10] [LZ93] [Man99] [Mar04] [Mas58]
[May72] [Mer04] [Mer05] [Mer06a] [Mer06b] [Mer09] [Mer10] [Mil10] [Mil11] [ML65] [ML95] [ML98] [MR96] [MS02]
[MS11] [MT11]
T. V. Kadeishvili, The algebraic structure in the homology of an A()-algebra, Soobshch. Akad. Nauk Gruzin. SSR 108 (1982), no. 2, 249252 (1983). 6 Ralph M. Kaufmann, On spineless cacti, Delignes conjecture and Connes-Kreimers Hopf algebra, Topology 46 (2007), no. 1, 3988. 38 R. M. Kaufmann, A proof of a cyclic version of Delignes conjecture via Cacti, Mathematical Research Letters 15 (2008), no. 5, 901921. 38 M. Kontsevich and Yu. Manin, Gromov-Witten classes, quantum cohomology, and enumerative geometry, Comm. Math. Phys. 164 (1994), no. 3, 525562. 17 Joachim Kock, Frobenius algebras and 2D topological quantum eld theories, London Mathematical Society Student Texts, vol. 59, Cambridge University Press, Cambridge, 2004. 16, 25 Maxim Kontsevich, Homological algebra of mirror symmetry, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Z urich, 1994) (Basel), Birkh auser, 1995, pp. 120139. 39 , Deformation quantization of Poisson manifolds, Lett. Math. Phys. 66 (2003), no. 3, 157216. 38 Jean-Louis Koszul, Homologie et cohomologie des alg` ebres de Lie, Bull. Soc. Math. France 78 (1950), 65127. 22 Maxim Kontsevich and Yan Soibelman, Deformations of algebras over operads and the Deligne conjecture, Conf erence Mosh e Flato 1999, Vol. I (Dijon), Math. Phys. Stud., vol. 21, Kluwer Acad. Publ., Dordrecht, 2000, pp. 255307. 38 , Homological mirror symmetry and torus brations, Symplectic geometry and mirror symmetry (Seoul, 2000), World Sci. Publ., River Edge, NJ, 2001, pp. 203263. 7 Hiroshige Kajiura and Jim Stasheff, Homotopy algebras inspired by classical open-closed string eld theory, Comm. Math. Phys. 263 (2006), no. 3, 553581. 39 M. Kontsevich and Y. Soibelman, Notes on A -algebras, A -categories and non-commutative geometry, Homological mirror symmetry, Lecture Notes in Phys., vol. 757, Springer, Berlin, 2009, pp. 153219. 38 F. William Lawvere, Functorial semantics of algebraic theories, Proc. Nat. Acad. Sci. U.S.A. 50 (1963), 869872. 18 K. Lefevre-Hasegawa, Sur les A-inni cat egories, arXiv.org:math/0310337 (2003). 7, 8 J.-L. Loday, Cyclic homology, second ed., Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 301, Springer-Verlag, Berlin, 1998, Appendix E by Mar a O. Ronco, Chapter 13 by the author in collaboration with Teimuraz Pirashvili. 38 Jean-Louis Loday and Daniel Quillen, Cyclic homology and the Lie algebra homology of matrices, Comment. Math. Helv. 59 (1984), no. 4, 569591. 38 Jean-Louis Loday and Bruno Vallette, Algebraic operads, to appear, 2010. 2, 19, 25, 26, 31, 34, 36, 40 Bong H. Lian and Gregg J. Zuckerman, New perspectives on the BRST-algebraic structure of string theory, Comm. Math. Phys. 154 (1993), no. 3, 613646. 38 Yuri I. Manin, Frobenius manifolds, quantum cohomology, and moduli spaces, American Mathematical Society Colloquium Publications, vol. 47, American Mathematical Society, Providence, RI, 1999. 17, 39 Martin Markl, Homotopy algebras are homotopy algebras, Forum Math. 16 (2004), no. 1, 129160. 7 W. S. Massey, Some higher order cohomology operations, Symposium internacional de topolog a algebraica International symposium on algebraic topology, Universidad Nacional Aut onoma de M exico and UNESCO, Mexico City, 1958, pp. 145154. 5 J.P. May, The geometry of iterated loop spaces, Springer-Verlag, Berlin, 1972, Lectures Notes in Mathematics, Vol. 271. 12, 14 S. A. Merkulov, Operads, deformation theory and F -manifolds, Frobenius manifolds, Aspects Math., E36, Vieweg, Wiesbaden, 2004, pp. 213251. 39 , Nijenhuis innity and contractible differential graded manifolds, Compos. Math. 141 (2005), no. 5, 12381254. 39 , PROP prole of Poisson geometry, Comm. Math. Phys. 262 (2006), no. 1, 117135. 39 S. A. Merkulov, Quantization of strongly homotopy Lie bialgebras, ArXiv Mathematics e-prints (2006). 27 S. A. Merkulov, Graph complexes with loops and wheels, Algebra, arithmetic, and geometry: in honor of Yu. I. Manin. Vol. II, Progr. Math., vol. 270, 2009, pp. 311354. 18 , Wheeled Pro(p)le of Batalin-Vilkovisky formalism, Comm. Math. Phys. 295 (2010), no. 3, 585638. 39 D. V. Millionshchikov, Massey products in Lie algebra cohomology. 37 M. Miller, Minimal Model for the Frobenius Algebra Structure of Differential Forms, ArXiv e-prints (2011). 39 Saunders Mac Lane, Categorical algebra, Bull. Amer. Math. Soc. 71 (1965), 40106. 17 , Homology, Classics in Mathematics, Springer-Verlag, Berlin, 1995, Reprint of the 1975 edition. 32 , Categories for the working mathematician, second ed., Graduate Texts in Mathematics, vol. 5, Springer-Verlag, New York, 1998. 32 Guy Melanc on and Christophe Reutenauer, Free Lie superalgebras, trees and chains of partitions, J. Algebraic Combin. 5 (1996), no. 4, 337351. 12 James E. McClure and Jeffrey H. Smith, A solution of Delignes Hochschild cohomology conjecture, Recent progress in homotopy theory (Baltimore, MD, 2000), Contemp. Math., vol. 293, Amer. Math. Soc., Providence, RI, 2002, pp. 153 193. 38 K. Muenster and I. Sachs, Quantum Open-Closed Homotopy Algebra and String Field Theory, ArXiv e-prints (2011). 39 F. Muro and A. Tonks, Unital associahedra, ArXiv:1110.1959 (2011). 29
43
[Mun78] [NR66] [NR67] [Pos93] [PP05] [Pri70] [Qui67] [Qui69] [Ree58] [RW98] [RW02] [Seg04] [Shi62] [ST09] [Sta63] [Ste47] [Sul77] [Sul10] [Tam51] [Tam98] [Tat57] [Tou06] [TTW11] [TZ06] [Val07] [Val11a] [Val11b] [VdL03] [Vor00] [Wei95] [Wil07] [WZ92] [Zwi93]
Hans J. Munkholm, DGA algebras as a Quillen model category. Relations to shm maps, J. Pure Appl. Algebra 13 (1978), no. 3, 221232. 8 Albert Nijenhuis and R. W. Richardson, Jr., Cohomology and deformations in graded Lie algebras, Bull. Amer. Math. Soc. 72 (1966), 129. 34 A. Nijenhuis and R. W. Richardson, Jr., Deformations of Lie algebra structures, J. Math. Mech. 17 (1967), 89105. 34 L. E. Positsel ski , Nonhomogeneous quadratic duality and curvature, Funktsional. Anal. i Prilozhen. 27 (1993), no. 3, 5766, 96. 28 A. Polishchuk and L. Positselski, Quadratic algebras, University Lecture Series, vol. 37, American Mathematical Society, Providence, RI, 2005. 28 S.B. Priddy, Koszul resolutions, Trans. Amer. Math. Soc. 152 (1970), 3960. 28 Daniel G. Quillen, Homotopical algebra, Lecture Notes in Mathematics, No. 43, Springer-Verlag, Berlin, 1967. 22 D. Quillen, Rational homotopy theory, Ann. of Math. (2) 90 (1969), 205295. 38 R. Ree, Lie elements and an algebra associated with shufes, Ann. of Math. (2) 68 (1958), 210220. 27 Dmitry Roytenberg and Alan Weinstein, Courant algebroids and strongly homotopy Lie algebras, Lett. Math. Phys. 46 (1998), no. 1, 8193. 38 Alan Robinson and Sarah Whitehouse, Operads and -homology of commutative rings, Math. Proc. Cambridge Philos. Soc. 132 (2002), no. 2, 197234. 32 G. Segal, The denition of conformal eld theory, Topology, geometry and quantum eld theory, London Math. Soc. Lecture Note Ser., vol. 308, Cambridge Univ. Press, Cambridge, 2004, pp. 421577. 17, 39 Weishu Shih, Homologie des espaces br es, Inst. Hautes Etudes Sci. Publ. Math. (1962), no. 13, 88. 36 Paolo Salvatore and Roberto Tauraso, The operad Lie is free, J. Pure Appl. Algebra 213 (2009), no. 2, 224230. 32 James Dillon Stasheff, Homotopy associativity of H -spaces. I, II, Trans. Amer. Math. Soc. 108 (1963), 275-292; ibid. 108 (1963), 293312. 4, 21 N. E. Steenrod, Products of cocycles and extensions of mappings, Ann. of Math. (2) 48 (1947), 290320. 9 D. Sullivan, Innitesimal computations in topology, Inst. Hautes Etudes Sci. Publ. Math. (1977), no. 47, 269331 (1978). 22, 38 D. Sullivan, Algebra, Topology and Algebraic Topology of 3D Ideal Fluids, ArXiv e-prints (2010). 39 D. Tamari, Monoides pr eordonn es et cha nes de malcev, Th` ese de Math ematique, Paris (1951). 21 Dmitry E. Tamarkin, Another proof of M. Kontsevich formality theorem, Doctoral thesis, Penn State, arXiv.org:math/9803025 (1998). 38 John Tate, Homology of Noetherian rings and local rings, Illinois J. Math. 1 (1957), 1427. 22 Victor Tourtchine, Dyer-Lashof-Cohen operations in Hochschild cohomology, Algebr. Geom. Topol. 6 (2006), 875894 (electronic). 32 J. Terilla, T. Tradler, and S. O. Wilson, Homotopy DG algebras induce homotopy BV algebras, ArXiv e-prints (2011). 38 T. Tradler and M. Zeinalian, On the cyclic Deligne conjecture, J. Pure Appl. Algebra 204 (2006), no. 2, 280299. 38 B. Vallette, A Koszul duality for props, Trans. of Amer. Math. Soc. 359 (2007), 48654993. 23, 27, 36 , Homotopy theory of homotopy algebras, in preparation (2011). 37 , Transfer of homotopy algebraic structures, in preparation (2011). 36 P. Van der Laan, Coloured Koszul duality and strongly homotopy operads, arXiv:math.QA/0312147 (2003). 35 Alexander A. Voronov, Homotopy Gerstenhaber algebras, Conf erence Mosh e Flato 1999, Vol. II (Dijon), Math. Phys. Stud., vol. 22, Kluwer Acad. Publ., Dordrecht, 2000, pp. 307331. 38 Alan Weinstein, Deformation quantization, Ast erisque (1995), no. 227, Exp. No. 789, 5, 389409, S eminaire Bourbaki, Vol. 1993/94. 38 S. O. Wilson, Free Frobenius algebra on the differential forms of a manifold, arXiv.org:0710.3550 (2007). 39 Edward Witten and Barton Zwiebach, Algebraic structures and differential geometry in two-dimensional string theory, Nuclear Phys. B 377 (1992), no. 1-2, 55112. 39 Barton Zwiebach, Closed string eld theory: quantum action and the Batalin-Vilkovisky master equation, Nuclear Phys. B 390 (1993), no. 1, 33152. 39
, U NIVERSIT E DE N ICE S OPHIA -A NTIPOLIS , PARC VALROSE , 06108 N ICE C EDEX 02, F RANCE L ABORATOIRE J.A. D IEUDONN E E-mail address: brunov@unice.fr

Algebra Homotopy Operad

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Algebra Homotopy Operad

Transféré par

Droits d'auteur :

Formats disponibles

ALGEBRA + HOMOTOPY = OPERAD